Live recordThe legal AI arena6,571 counted voices14 tools ranked · 97 in the marketUpdated Oct 26, 2026
Home › Arena
+37 net · 872 voices · High confidence
From 6,571 lawyers’ own posts that state an experience or a verdict, ranked by how sure we can be, not by the loudest. For contracts lawyers name Claude; for research, Perplexity and Clio are tied. Below: every task, every tool, every model, and what outside benchmarks say.
Best overall · 60+ voices
90% band +33 to +41 · Tied with Spellbook, Gemini (bands overlap).
As a lawyer, Claude 4 is the best writer, and usually, but not always, the leader in legal reasoning.
Hacker News · Jun 5, 2025
Frontier today DeepSeek V4.1 Flash 2026-10-17 · Gemini 4 Argon 2026-10-13 · Muse Wonder Full 2026-10-12 · Mistral Large 4 2026-10-06 · GPT-6 Astra 2026-09-03 · Claude Fable 5.1 2026-09-01 · Grok 4.7 2026-09 (month only) · Kimi K3 date not published · Qwen3.8 Max date not published · Command A+ date not published All models and scores →
Cite this: “As of Oct 26, 2026, Claude has the highest floor of lawyer sentiment among AI tools with 60+ voices on aiforlawyers.app (net +37, 90% band +33 to +41, 872 voices).” How it is ranked
Every tool with 10+ voices. Dot = net (like minus dislike, per 100 voices); bar = the 90% band, how far net could move with a different sample. Order follows the floor of the band, so 27 happy voices do not outrank 541. Best overall needs 60+ voices.
Green: the whole band is above zero. Red: below zero. Gold: crosses zero, not settled. Source: 6,571 counted voices, Oct 26, 2026. The full table →
The newest top model from each lab (its flagship where the lab names one), with the release date from the lab’s own page (checked 2026-10-23), and what public benchmarks measure today. These are tests on question sets, not lawyers’ experience; our lawyer-voice ranking is the one above and never mixes with these.
Dated report: Frontier models for legal work, Claude vs ChatGPT vs Gemini, our record and public benchmarks kept apart (PDF).
Public benchmarks, dated and summarised: Legal AI benchmarks (Vals VLAIR, Stanford hallucination study, LegalBench), other publishers’ figures kept apart from ours (PDF).
Each model works through US legal research questions as an agent; a question counts only if every item of the grading rubric passes. Results page, read 2026-10-23.
Scale 0 to 60%. Ties at the top are real ties.
| Model | Released | Legal Research Bench | LegalBench | AA Intelligence | Arena rank | Harvey LAB | Lawyers name it | |
|---|---|---|---|---|---|---|---|---|
| DeepSeek V4.1 Flash flagshipDeepSeek | 2026-10-17 | 45.19% | – | 45 | #39 | 1.7% | 1 voice | source |
| Gemini 4 Argon flagshipGoogle | 2026-10-13 | 54.81% | 88.3% | 53 | #1 | 3.9% | none yet | source |
| Muse Wonder Full flagshipMeta | 2026-10-12 | 53.37% | 87.42% | 56 | – | 13.1% | none yet | source |
| Muse WonderMeta | 2026-10-12 | 50.48% | – | 52 | – | 10% | none yet | source |
| Claude Haiku 5.5Anthropic | 2026-10-07 | 43.27% | – | 43 | – | 2.8% | none yet | source |
| Mistral Large 4 flagshipMistral | 2026-10-06 | 31.73% | – | 38 | #115 | 3.1% | none yet | source |
| GPT-6.1 SolOpenAI | 2026-09-29 | 38.46% | – | 52 | #21 | 6.9% | none yet | source |
| Claude Sonnet 5.5Anthropic | 2026-09-28 | 48.08% | – | 56 | #32 | 2.8% | none yet | source |
| GPT-6 SolOpenAI | 2026-09-22 | 28.85% | – | – | #67 | 3.6% | none yet | source |
| GPT-6 LunaOpenAI | 2026-09-22 | 30.29% | – | 38 | #91 | 3.3% | none yet | source |
| Claude Opus 5.5Anthropic | 2026-09-22 | 50.48% | – | 58 | #2 | 4.2% | 4 voices | source |
| GPT-6 Astra flagshipOpenAI | 2026-09-03 | 39.42% | – | 53 | #35 | 8.6% | 2 voices | source |
| Gemini 3.8 FlashGoogle | 2026-09-02 | 38.94% | 86.99% | 41 | #8 | 0% | none yet | source |
| Muse Spark 1.3Meta | 2026-09-02 | 55.29% | – | – | #9 | 8.9% | 1 voice | source |
| Claude Fable 5.1 flagshipAnthropic | 2026-09-01 | 55.29% | 88.51% | 53 | #6 | 6.4% | 2 voices | source |
| Grok 4.7 flagshipxAI | 2026-09 (month only) | 47.12% | 84.39% | 46 | #90 | 9.4% | none yet | source |
| DeepSeek V4-ProDeepSeek | 2026-08-13 | 40.87% | 82.36% | 36 | #55 | – | none yet | source |
| Grok 4.6xAI | 2026-08 (month only) | 48.08% | 86.31% | – | #72 | – | 1 voice | source |
| Claude Opus 5Anthropic | 2026-07-24 | 55.29% | 86.97% | – | #13 | – | 10 voices | source |
| GPT-5.6 SolOpenAI | 2026-07-09 | 48.08% | 86.97% | – | #20 | – | 5 voices | source |
| Grok 4.5xAI | 2026-07 (month only) | 37.98% | 85.97% | – | #51 | – | 2 voices | source |
| Claude Sonnet 5Anthropic | 2026-06-30 | 41.83% | 83.92% | – | #57 | – | 2 voices | source |
| Claude Fable 5Anthropic | 2026-06-09 | 49.52% | 88.56% | – | #4 | – | 10 voices | source |
| Mistral Medium 3.5Mistral | 2026-04-28 | 9.13% | – | 14 | #117 | – | none yet | source |
| Gemini 3.1 ProGoogle | 2026-02-19 | 20.67% | 87.4% | 30 | #17 | 0% | 4 voices | source |
| Mistral Large 3Mistral | 2025-12-02 | – | 79.14% | – | #143 | – | none yet | source |
| DeepSeek V3.2DeepSeek | 2025-12-01 | – | 76.08% | – | #122 | – | none yet | source |
| Llama 4Meta | date not published | – | – | – | – | – | 1 voice | source |
| Kimi K3 flagshipMoonshot | date not published | 46.15% | 86.21% | 44 | #16 | 5.3% | 6 voices | source |
| Qwen3.8 Max flagshipAlibaba | date not published | 47.6% | 83.61% | – | #22 | – | none yet | source |
| Command A+ flagshipCohere | date not published | – | 61.42% | 13 | – | – | none yet | source |
A dash means the benchmark did not list that model when we read it; we never estimate a missing score. Benchmarks: Vals Legal Research Bench (All-Pass accuracy) (updated 2026-10-22, read 2026-10-23) · Vals LegalBench (updated 2026-10-13, read 2026-10-23) · Artificial Analysis Intelligence Index v4.3.2 (read 2026-10-23) · Arena (formerly LMArena) Text, Overall, default view (updated 2026-10-08, read 2026-10-23, from the Internet Archive copy of 2026-10-12 (the live page blocks automated reads)) · Harvey LAB-AA (Artificial Analysis), hallucination-gated all-pass rate (read 2026-10-23).
Chat answers a question; an agent takes a task and works through the steps: reads the data room, runs the review, drafts, checks, hands back a result. 33 legal AI products that ship agents today, each with what it executes in the vendor’s own words and a link (read 2026-10-23). 3 say they run a workflow end to end; the rest keep a lawyer at checkpoints. Being listed here is not a rating: what lawyers say about each tool is in the ranking above.
AI voice and SMS agents handle routine case communications for personal injury firms (EvenUp launch post).
Launched 2026-01-28 · vendor source
Flankruns end to endFlank agents handle high-volume contracting workflows end-to-end; lawyers check the finished work and high-risk moments (Flank homepage).
Luminanceruns end to endCombines its legal LLM with AI tools and memory to create an active agent that performs tasks and makes decisions without needing human oversight (Luminance press release, 2024-10-22).
Launched 2024-10-22 · vendor source
Genius Mode is a capability within August's Assistant that handles multi-step, multi-document tasks end to end: it reads a full set of inputs, decides what needs to be produced, and assembles everything, flagging open issues (August blog).
Para is an AI colleague for legal teams: tell Para what needs to get done, it does the work, keeps going after you close your laptop, and checks with you before anything leaves the firm (CaseMark homepage).
Launched 2026-10-19 · vendor source
Next-generation CoCounsel Legal is engineered to reason, plan, and execute at the level of a senior associate, running multi-step legal workflows rather than answering isolated prompts (TR press release, 2026-08-20). Deep Research generates multi-step research plans and delivers citation-backed reports (2025-08-05 release).
Launched 2025-08-05 · vendor source
Crosbymulti-step with human checkpointsAI agents research, analyze, and draft on your behalf while human attorneys review all work, manage agents, and ensure quality (Crosby homepage).
Agents take on multi-step legal work and return it for your approval (Eudia homepage).
Evemulti-step with human checkpointsAgents monitor cases and act when something needs to move forward: when records arrive they summarize them, when deadlines approach they draft, when clients need updates they handle communication (Eve 2.0 launch post).
Launched 2026-01-13 · vendor source
Filevine AI (LOIS)multi-step with human checkpointsLOIS writes results back into the firm's system of record and then executes actions on the firm's behalf (Filevine press release).
Launched 2026-06-02 · vendor source
Agent Connectors put GC AI to work in the tools you already use: reading your email, files, calendar, and Slack, and taking action in them with your approval; contract agents in Word redline and act on your behalf (GC AI homepage).
Harveymulti-step with human checkpointsPurpose built agents execute complex legal work end to end (Harvey platform page, quoted in the launch post).
Launched 2026-05-05 · vendor source
Jylomulti-step with human checkpointsCapture your expertise and run it as playbooks or agents; the demo shows it spawning one agent per lease and bringing findings back as one cited table (Jylo homepage).
Agent Studio automates the entire workflow, built around your organisation's logic, playbooks and approval flows; workflows can be triggered from email, Slack or Teams (Legalfly April 2026 product highlights).
Launched 2026-04-09 · vendor source
LegalOnmulti-step with human checkpointsLegalOn's agentic AI plans, reasons and executes multi-step assignments within common legal workflows, while keeping human expertise firmly in the loop (LegalOn press release).
Launched 2026-02-10 · vendor source
An Orchestrator Agent breaks down complex or lengthy legal queries into manageable tasks and coordinates multiple agents working together; a Reflection Agent reviews and strengthens the final response (LexisNexis press release).
Launched 2025-08-11 · vendor source
Norm Aimulti-step with human checkpointsConverts regulations into computer programs that can make autonomous compliance determinations, so compliance teams delegate tasks to Regulatory AI via high-level instructions and validate the work product (Norm Ai article).
Relativity aiRmulti-step with human checkpointsThe systems engage in multi-step reasoning and act by making and applying informed coding decisions on documents; Case Strategy extracts facts, drafts summaries and builds chronologies (Relativity blog).
multi-step with human checkpointsDelegate full multi-step projects: it breaks the project into a plan, executes it, adjusts the plan if things go wrong, works across multiple documents and data sources, revises Word documents, asks for help, and checks its own work (Spellbook launch post).
Launched 2024-08-22 · vendor source
Supiomulti-step with human checkpointsAgentic AI built for plaintiff attorneys: drafting demands, tracking liens, scoring leads, and turning institutional knowledge into case wins (Supio press release).
Launched 2026-05-14 · vendor source
vLex Vincentmulti-step with human checkpointsAgentic Mode lets you describe the outcome you need, and Vincent plans and executes the full workflow, research, analysis, and drafting, in one result (vLex knowledge base).
Wordsmithmulti-step with human checkpointsAgents are AI workers that receive work from the business, execute multi-step workflows within your standards, and deliver the outcome (Wordsmith launch post).
Launched 2026-05-14 · vendor source
Alexiautonomy not statedAutomate what's repeatable by building or customizing workflows, using pre-built templates or firm-specific agents to streamline production of research memos, drafts, and reviews (Alexi AI engine page).
Chamelioautonomy not statedChamelio's agentic workflows execute complex, multi-step tasks across the entire legal lifecycle, from routing requests to surgical redlining and liability ranking (Chamelio homepage).
DeepJudgeautonomy not statedBuild, deploy and orchestrate AI agents powered by your firm data; its agentic reasoning executes complex, multi-step tasks, retrieving data from multiple sources and taking actions (DeepJudge homepage).
Definelyautonomy not statedDefinely's agentic system analyses the task, creates a plan, and assigns agents to handle different parts of the task; agents collaborate and pass information along until completion (Definely launch post).
Launched 2025-05-22 · vendor source
DISCOautonomy not statedAdds an autonomous, multi-step reasoning engine to its Cecilia Q&A tool that returns more detailed results on very large data sets (DISCO press release).
Launched 2026-02-09 · vendor source
DraftWiseautonomy not statedAn AI agent that drafts, reviews, and negotiates contracts using your firm's own precedent and language (Draftwise page description).
Ironclad Agent is the front door to the AI platform: one conversational interface to access contracts and workflows and orchestrate specialized AI agents; it calls Contract Review Agent to redline contracts against approved positions and deal history (Ironclad press release).
Launched 2026-10-08 · vendor source
Juroautonomy not statedAn agentic AI product that enables business users to review and redline contracts automatically against custom playbooks (Juro launch post).
Launched 2025-07-25 · vendor source
Lito is Litera's new AI legal agent, included automatically in its drafting products and in Kira at no additional charge (Litera press release, 2025-10-27).
Launched 2025-10-27 · vendor source
Legoraautonomy not statedThe Agent plans, executes, reviews and delivers end-to-end execution of complex legal work (Legora homepage); press release: a purpose-built agentic operating system that enables legal teams to execute complex legal work end-to-end.
Launched 2026-05-07 · vendor source
autonomy not statedBreaks down difficult tasks into multi-step AI workflows that mirror how legal professionals operate, improving the quality and depth of document creation (Paxton post).
Net on each task, with the number of voices under it. A cell needs 10+ voices on that task. Outlined: the leader of the task by band floor.
| Tool | Research | Drafting | Contracts | Citations | Discovery | Summaries | General |
|---|---|---|---|---|---|---|---|
 ChatGPT | -34623 | +22573 | -1157 | -69157 | +932 | +5114 | -18790 |
| Claude | +4182 | +57141 | +4665 | +1631 | +3517 | +5927 | +28455 |
| Harvey | -1156 | +2059 | +1243 | – | +2010 | +1119 | -30545 |
| CoCounsel | +17168 | +335 | – | – | +018 | -3033 | -3135 |
| Lexis+ AI | -10184 | -1118 | – | -4717 | – | -4010 | -1061 |
| Westlaw AI | +13210 | – | – | -4511 | – | -911 | -2931 |
 Copilot | -3010 | +1730 | – | – | – | +1119 | +078 |
| Gemini | +4231 | +4827 | – | – | – | +3611 | +1145 |
| – | +3222 | +3951 | – | – | – | +4323 |
| Legora | – | – | – | – | – | – | -563 |
 Clio | +6411 | – | – | – | – | – | -2434 |
| Perplexity | +8817 | – | – | – | – | – | – |
| vLex Vincent | +5721 | – | – | – | – | – | – |
The pairs people search for most (Google Search Console impressions on our sister site aivortex.io, Jul 21 to Oct 18, 2026), each side with its record. Every comparison →
1,330 search impressions for this pair
| ChatGPT | Claude | |
|---|---|---|
| Voices | 2,467 | 872 |
| Net | -13 | +37 |
| 90% band | -16 to -10 | +33 to +41 |
| Top complaint | Hallucination | Privacy |
Claude ahead; the 90% bands do not overlap. Full comparison →
1,127 search impressions for this pair
| Claude | Harvey | |
|---|---|---|
| Voices | 872 | 749 |
| Net | +37 | -21 |
| 90% band | +33 to +41 | -26 to -15 |
| Top complaint | Privacy | Accuracy |
Claude ahead; the 90% bands do not overlap. Full comparison →
601 search impressions for this pair
| Harvey | Spellbook | |
|---|---|---|
| Voices | 749 | 99 |
| Net | -21 | +40 |
| 90% band | -26 to -15 | +27 to +52 |
| Top complaint | Accuracy | Accuracy |
Spellbook ahead; the 90% bands do not overlap. Full comparison →
506 search impressions for this pair
| Lexis+ AI | Westlaw AI | |
|---|---|---|
| Voices | 294 | 268 |
| Net | -14 | +6 |
| 90% band | -23 to -6 | -3 to +14 |
| Top complaint | Accuracy | Accuracy |
Westlaw AI ahead; the 90% bands do not overlap. Full comparison →
479 search impressions for this pair
| CoCounsel | Harvey | |
|---|---|---|
| Voices | 411 | 749 |
| Net | +2 | -21 |
| 90% band | -4 to +9 | -26 to -15 |
| Top complaint | Accuracy | Accuracy |
CoCounsel ahead; the 90% bands do not overlap. Full comparison →
418 search impressions for this pair
| Harvey | Legora | |
|---|---|---|
| Voices | 749 | 90 |
| Net | -21 | -3 |
| 90% band | -26 to -15 | -18 to +11 |
| Top complaint | Accuracy | Accuracy |
Bands overlap: on this record the order is not settled. Full comparison →
417 search impressions for this pair
| Harvey | Lexis+ AI | |
|---|---|---|
| Voices | 749 | 294 |
| Net | -21 | -14 |
| 90% band | -26 to -15 | -23 to -6 |
| Top complaint | Accuracy | Accuracy |
Bands overlap: on this record the order is not settled. Full comparison →
257 search impressions for this pair
| Westlaw AI | vLex Vincent | |
|---|---|---|
| Voices | 268 | 31 |
| Net | +6 | +52 |
| 90% band | -3 to +14 | +24 to +70 |
| Top complaint | Accuracy | Accuracy |
vLex Vincent ahead; the 90% bands do not overlap. Full comparison →
For each task, every tool with 10+ voices on it, ranked by the floor of the 90% band. When two bands overlap we say the order is not settled.
Perplexity (+88) and Clio (+64) have overlapping 90% bands, so the order between them is not settled. What lawyers say about legal research →
| # | Tool | Voices | Net 90% band | Confidence | Reading |
|---|---|---|---|---|---|
| 1 | Perplexity | 17 | +88 +62 to +94 | Low | leads on the floor of its band |
| 2 | Clio | 11 | +64 +30 to +80 | Low | Tied with #1 |
| 3 | Claude | 82 | +41 +27 to +54 | Medium | below #1, bands apart |
| 4 | vLex Vincent | 21 | +57 +26 to +76 | Low | Tied with #1 |
| 5 | Gemini | 31 | +42 +19 to +59 | Low | below #1, bands apart |
| 6 | CoCounsel | 168 | +17 +6 to +27 | Medium | below #1, bands apart |
| 7 | Westlaw AI | 210 | +13 +4 to +22 | Medium | below #1, bands apart |
| 8 | Lexis+ AI | 184 | −10 −21 to 0 | Medium | below #1, bands apart |
| 9 | Harvey | 56 | −11 −27 to +6 | Low | below #1, bands apart |
| 10 | ChatGPT | 623 | −34 −39 to −28 | High | below #1, bands apart |
| 11 | Copilot | 10 | −30 −66 to +18 | Low | below #1, bands apart |
Claude (+57) and Gemini (+48) have overlapping 90% bands, so the order between them is not settled. What lawyers say about drafting →
| # | Tool | Voices | Net 90% band | Confidence | Reading |
|---|---|---|---|---|---|
| 1 | Claude | 141 | +57 +47 to +65 | Medium | leads on the floor of its band |
| 2 | Gemini | 27 | +48 +20 to +68 | Low | Tied with #1 |
| 3 | ChatGPT | 573 | +22 +16 to +28 | High | below #1, bands apart |
| 4 | | 22 | +32 +5 to +53 | Low | Tied with #1 |
| 5 | Harvey | 59 | +20 +3 to +36 | Low | below #1, bands apart |
| 6 | Copilot | 30 | +17 −8 to +40 | Low | below #1, bands apart |
| 7 | CoCounsel | 35 | +3 −22 to +27 | Low | below #1, bands apart |
| 8 | Lexis+ AI | 18 | −11 −42 to +22 | Low | below #1, bands apart |
Claude (+46) and Spellbook (+39) have overlapping 90% bands, so the order between them is not settled. What lawyers say about contract review →
Claude (+16) and Lexis+ AI (−47) have overlapping 90% bands, so the order between them is not settled. What lawyers say about citation checking →
| # | Tool | Voices | Net 90% band | Confidence | Reading |
|---|---|---|---|---|---|
| 1 | Claude | 31 | +16 −12 to +41 | Low | leads on the floor of its band |
| 2 | Lexis+ AI | 17 | −47 −71 to −9 | Low | Tied with #1 |
| 3 | ChatGPT | 157 | −69 −76 to −59 | Medium | below #1, bands apart |
| 4 | Westlaw AI | 11 | −45 −77 to +3 | Low | Tied with #1 |
Claude (+35) and ChatGPT (+9) have overlapping 90% bands, so the order between them is not settled. What lawyers say about discovery →
Claude (+59) is above ChatGPT (+5) and the two 90% bands do not overlap. What lawyers say about summarization →
| # | Tool | Voices | Net 90% band | Confidence | Reading |
|---|---|---|---|---|---|
| 1 | Claude | 27 | +59 +33 to +75 | Low | leads on the floor of its band |
| 2 | ChatGPT | 114 | +5 −8 to +18 | Medium | below #1, bands apart |
| 3 | Gemini | 11 | +36 −10 to +68 | Low | Tied with #1 |
| 4 | Copilot | 19 | +11 −19 to +38 | Low | Tied with #1 |
| 5 | Harvey | 19 | +11 −24 to +41 | Low | Tied with #1 |
| 6 | Westlaw AI | 11 | −9 −43 to +29 | Low | below #1, bands apart |
| 7 | CoCounsel | 33 | −30 −49 to −7 | Low | below #1, bands apart |
| 8 | Lexis+ AI | 10 | −40 −70 to +5 | Low | below #1, bands apart |
Claude (+28) and Spellbook (+43) have overlapping 90% bands, so the order between them is not settled.
| # | Tool | Voices | Net 90% band | Confidence | Reading |
|---|---|---|---|---|---|
| 1 | Claude | 455 | +28 +22 to +34 | High | leads on the floor of its band |
| 2 | | 23 | +43 +12 to +65 | Low | Tied with #1 |
| 3 | Gemini | 45 | +11 −11 to +33 | Low | Tied with #1 |
| 4 | CoCounsel | 135 | −3 −15 to +9 | Medium | below #1, bands apart |
| 5 | Copilot | 78 | 0 −16 to +16 | Medium | below #1, bands apart |
| 6 | ChatGPT | 790 | −18 −23 to −13 | High | below #1, bands apart |
| 7 | Legora | 63 | −5 −23 to +14 | Medium | below #1, bands apart |
| 8 | Lexis+ AI | 61 | −10 −28 to +9 | Medium | below #1, bands apart |
| 9 | Harvey | 545 | −30 −36 to −24 | High | below #1, bands apart |
| 10 | Clio | 34 | −24 −46 to +3 | Low | below #1, bands apart |
| 11 | Westlaw AI | 31 | −29 −52 to −2 | Low | below #1, bands apart |
Models and versions lawyers name in their own words, with 10+ voices that name them, ranked by the floor of the 90% band. Best named version per family: Claude 3.5 Sonnet +100 (10 voices) · Gemini 3 +80 (10 voices) · o3 +82 (11 voices) · CoCounsel (Casetext era) +33 (75 voices) · Lexis+ AI Protégé +6 (65 voices) · Westlaw Precision 0 (18 voices). By model →
| # | Model | Voices | Net 90% band | Confidence | Dislike per 100 |
|---|---|---|---|---|---|
| 1 | Claude 3.5 SonnetClaude, 2024 to 2026 | 10 | +100 +61 to +99 | Low | 0 |
| 2 | Gemini 3Gemini, 2025 to 2026 | 10 | +80 +41 to +91 | Low | 0 |
| 3 | o3ChatGPT, 2025 to 2025 | 11 | +82 +37 to +95 | Low | 9 |
| 4 | Gemini 2.xGemini, 2025 to 2026 | 14 | +57 +22 to +76 | Low | 7 |
| 5 | CoCounsel (Casetext era)CoCounsel, 2023 to 2026 | 75 | +33 +17 to +47 | Medium | 19 |
| 6 | GPT-4oChatGPT, 2024 to 2026 | 23 | +43 +13 to +65 | Low | 17 |
| 7 | Claude Opus 4.xClaude, 2025 to 2026 | 15 | +47 +12 to +70 | Low | 13 |
| 8 | GPT-4ChatGPT, 2023 to 2026 | 65 | +12 −4 to +28 | Medium | 28 |
| 9 | Lexis+ AI ProtégéLexis+ AI, 2025 to 2026 | 65 | +6 −11 to +23 | Medium | 35 |
| 10 | GPT-5ChatGPT, 2025 to 2026 | 23 | +13 −16 to +39 | Low | 30 |
| 11 | Westlaw PrecisionWestlaw AI, 2024 to 2026 | 18 | 0 −32 to +33 | Low | 39 |
| 12 | BardGemini, 2023 to 2025 | 11 | −36 −69 to +11 | Low | 64 |
The newest frontier models, with release dates and benchmark scores, are in Frontier models today.
Net by quarter for the five most discussed tools; a point needs 5+ voices that quarter. 2026-Q4 is partial. With release dates →
Words and pairs of words that set liked voices apart from disliked ones (words in 8+ voices, scored by how much more often they show up on one side than the other), then what sets each tool apart. Bigger = more distinctive. Tool names are left out.
ChatGPT 2,467 voicesHarvey 749 voicesPer tool, green words lean liked, red lean disliked, gold are split.
One voice per leading tool, verbatim, dated and linked, with the platform it came from. Questions without a position and text quoted from someone else are never shown as a voice.
“It saves so much time and finds cases others have not.”
“As a lawyer, Claude 4 is the best writer, and usually, but not always, the leader in legal reasoning.”
“Hope it doesn‘t show a bad customer service over all because the product itself is good…”
“It is an excellent tool that saves an enormous amount of time, but you have to know how to use it i.e. very targeted prompts and keywords.”
“Considering Opus 5.5 only gets a 4% on the test compared to Gemini’s 20%, that’s actually pretty incredible.”
“I feel bad about using it but it’s infinitely Faster. We use Paxton and it’s great”
What independent tests and leaderboards say, each with its publication date, the day we re-read it and a link. They measure answers to test questions; we record what practicing lawyers say after using the tools. The two are never mixed: nothing here changes a number above.
legal-specific vendor test · published 2025-02-27 · read Oct 22, 2026
Vals AI
Four tools were tested on seven tasks against a lawyer baseline: CoCounsel, Vincent AI, Harvey Assistant and Oliver (Lexis+ AI withdrew from the sections studied). Harvey took the top score on five of the six tasks it entered, including Document Q&A at 94.8% and Chronology Generation at 80.2%; CoCounsel scored 89.6% on Document Q&A and 77.2% on Document Summarization, the top score there. The lawyer baseline beat the AI tools on EDGAR research and on Redlining (79.7%).
Top: Harvey Assistant (top on 5 of the 6 tasks it entered); CoCounsel top on Document Summarization
Vendors opted in task by task (Harvey six of seven, CoCounsel four), so tools were not compared on the same tasks; 2025 versions, over a year old; not a ranking of all legal AI.
vals.ai/industry-reports/vlairlegal-specific vendor test · published 2025-10-14 · read Oct 22, 2026
Vals AI
Four products (Alexi, Counsel Stack, Midpage, ChatGPT) and a lawyer baseline answered 200 US legal research questions. Weighted scores were 74% to 78% for the AI products and 69% for the lawyers; the page says Counsel Stack received the highest score across all criteria, and that legal AI and generalist AI both scored about 80% on accuracy (78 to 81% legal AI, 80% ChatGPT).
Top: Counsel Stack
Only four products took part; the page lists Reed Smith, Fisher Philips, McDermott, Ogletree, Paul Hastings and Paul Weiss as the question-writing consortium. Trade press (legaltechnology.com headline, not re-read here) says the market leaders were absent. Lawyer baseline answers were shorter by default, which the report says lowered their scores.
vals.ai/industry-reports/vlair-10-14-25legal reasoning benchmark · published 2026-10-22 (page shows Updated 10/22/2026) · read Oct 22, 2026
Vals AI
As of October 22, 2026 the page reports Muse Spark 1.3 Max, Claude Opus 5 and Claude Fable 5.1 tied for first at 55.29% (all-pass accuracy). Next: Gemini 4 Argon 54.81%, Wonder Full 53.37%, Wonder and Claude Opus 5.5 both 50.48%, GPT-5.6 Sol 48.08%. Cost per test ranges from $0.59 (Muse Spark 1.3 Max) to $25.16 (Claude Opus 5.5).
Top: Muse Spark 1.3 Max, Claude Opus 5 and Claude Fable 5.1 (tie, 55.29%)
Tests raw models in an agent harness, not legal products like Harvey or CoCounsel; private test set; 'All-Pass' means every rubric item must pass. Top 3 are a three-way tie, so rank order is arbitrary.
vals.ai/benchmarks/legal_researchlegal reasoning benchmark · published 2026-10-13 (page shows Updated 10/13/2026) · read Oct 22, 2026
Vals AI
As of October 13, 2026: 1 Claude Fable 5 88.56%, 2 Claude Fable 5.1 88.51%, 3 Gemini 4 Argon 88.30%; Wonder Full 87.42%, Gemini 3.1 Pro Preview 87.40%, GPT-5.6 Sol 86.97%, Claude Opus 5 86.97%, Grok 4.6 86.31%. The whole top 10 sits within about 1.6 points.
Top: Claude Fable 5 (88.56%)
Mostly classification and rule-recall tasks on public data, so contamination is possible and the top scores are nearly saturated; measures models, not legal products.
vals.ai/benchmarks/legal_benchlegal reasoning benchmark · published 2023-08-20 · read Oct 22, 2026
Stanford / Hazy Research (40 authors)
The abstract describes 162 tasks covering six types of legal reasoning, hand-crafted by legal professionals, used to evaluate 20 LLMs. The project site (hazyresearch.stanford.edu/legalbench) hosts the task list and getting-started pages; it shows no current leaderboard, which is hosted by Vals (see entry above).
2023 paper; models evaluated are long superseded; the benchmark is public, so newer models may have seen it.
arxiv.org/abs/2308.11462hallucination study · published 2024-05-30 (arXiv v1; HAI news post dated 2024-05-23) · read Oct 22, 2026
Stanford RegLab and HAI
The preregistered study found Lexis+ AI and Thomson Reuters' Westlaw AI-Assisted Research and Ask Practical Law AI hallucinated 17% to 33% of the time per the arXiv abstract, fewer than GPT-4. The HAI post puts Lexis+ AI and Ask Practical Law AI at more than 17% and Westlaw AI-Assisted Research at more than 34%.
Tested 2024 versions; Westlaw range differs between abstract (up to 33%) and HAI post (above 34%); it covered only LexisNexis and Thomson Reuters products plus GPT-4 (no Harvey, Legora or Spellbook), and it ranks nothing overall.
arxiv.org/abs/2405.20362general model leaderboard · published 2026-10-15 (leaderboard date shown; archive snapshot 2026-10-17) · read Oct 22, 2026
LMArena
Live pages block automated fetch (403 Cloudflare). The Wayback copy captured 2026-10-17 of arena.ai/leaderboard/chat/text shows data dated Oct 15, 2026 (416 models, 8,834,211 votes): 1 Gemini 4 Argon (High) 1524 (marked Preliminary, 4,904 votes), 2 Muse Wonder (Max) 1509, 3 Claude Opus 5.5 (High) 1508; then Claude Fable 5 (High) 1504 and Claude Opus 4.6 (High) 1503. A snapshot from a week before today, not a live read.
Top: Gemini 4 Argon (High), 1524, preliminary
Read from the Internet Archive copy captured 2026-10-17: the live page blocks automated reads.
Preference votes on general chat, not legal accuracy; the top 3 are within overlapping uncertainty bands (±9, ±11, ±7); #1 is flagged preliminary; archive copy, not fetched live today.
web.archive.org/web/20261017054907/https://arena.ai/leaderbogeneral model leaderboard · published 2026-10-22 (changelog on the page lists entries through 22 Oct) · read Oct 22, 2026
Artificial Analysis
Top by index score: Claude Opus 5.5 (max with fallback) 58, Claude Sonnet 5.5 (max with fallback) 56, Claude Opus 5.5 (xhigh with fallback) 56 and Muse Wonder Full (max) 56; next Claude Fable 5.1 53, GPT-6 Astra (max) 53 and Gemini 4 Argon (high) 53. The index combines 10 evaluations (e.g. GDPval-AA, Terminal-Bench 4.0, Humanity's Last Exam, AA-Omniscience).
Top: Claude Opus 5.5 (max with fallback), 58
Mostly coding, agentic and science tasks, nothing legal; scores depend on reasoning-effort setting; the page text I read does not print an explicit as-of date other than its changelog.
artificialanalysis.ai/leaderboards/modelslegal-specific vendor test · published 2026-05-06 (launch per Harvey blog coverage); leaderboard date not shown on page · read Oct 22, 2026
Harvey (benchmark) / Mercor (leaderboard)
Open-source benchmark of 1,749 tasks across 24 practice areas plus in-house contracting, graded with all-pass rubrics (100,000+ criteria). Pass@1 top 5 on the page: Gemini 4 Argon High 27.4% (±5.4), Grok 4.7 xHigh 19.5% (±4.5), Muse Spark 1.2 xHigh 19.4% (±4.6), Claude Opus 5 Max 17.8% (±1.5), Claude Fable 5.1 Max 17.4% (±4.2).
Top: Gemini 4 Argon High (27.4%)
Built by Harvey (a vendor); the leaderboard lists foundation models, not Harvey's own product; leaderboard publication date not shown and the top-5 error bars overlap heavily; the leaderboard page counts 1,749 tasks versus 'more than 1,200' reported at launch, so the set has grown.
mercor.com/apex/partner-benchmarks/harvey-lab-leaderboard/legal-specific vendor test · published 2024-08-29 · read Oct 22, 2026
Harvey
Harvey says its proprietary assistant models produce 74% of a final expert-lawyer-quality work product, outperforming leading foundation models on answer scores; foundation models gave good answers but struggled to cite sources, and prompts forcing sources led them to hallucinate them.
Top: Harvey (self-reported)
Run by the vendor on its own dataset, 2024 models; read as marketing evidence, not an independent comparison.
harvey.ai/blog/introducing-biglaw-benchlegal reasoning benchmark · published 2024-08-19 · read Oct 22, 2026
arXiv preprint (Pipitone, Houir Alami)
Introduced as the first benchmark designed to evaluate the retrieval step of RAG pipelines in the legal domain, built by tracing LegalBench query contexts back to their original locations in a legal corpus, and emphasizing minimal, precise snippets. The abstract gives no model scores.
Measures retrieval only, not answer quality; no scores in the abstract; 2024 paper.
arxiv.org/abs/2408.10343legal reasoning benchmark · published 2026-05-04 (page shows Updated 5/4/2026) · read Oct 22, 2026
Vals AI
Archived because performance saturated and it is no longer run on new models. Final board: 1 Grok 4.3 79.31%, 2 GPT 5.1 73.42%, 3 GPT 4.1 69.88%; Claude Opus 4.7 68.38%.
Top: Grok 4.3 (79.31%, frozen as of 2026-05-04)
Canadian case law only, private questions, and the page says it is archived, so it says nothing about 2026-09 and later models.
vals.ai/benchmarks/case_law_v2drafting-specific independent test · published 2025-09 (preliminary findings; tools run July to August 2025) · read Oct 23, 2026
Legal Benchmarks (Anna Guo and co-authors)
30 contract-drafting tasks (basic clauses, template adaptation, bespoke clauses; junior to mid level) written by practising lawyers, run on 13 AI tools plus in-house commercial lawyers (about 10 years' experience) as a baseline, 450 task outputs in all. A draft passed only if it met every key element from the contributing lawyer's answer key plus instruction compliance, factual accuracy and legal adequacy. Pass rate averaged 57.3% across the AI tools; the human lawyers reached 56.7%. Top individual tool: Gemini 2.5 Pro at 73.3%, against 70% for the best individual lawyer. The authors report that legal-specific tools did not clearly beat general assistants on draft quality, and that legal tools raised explicit risk warnings more often (83% of high-risk outputs versus 55% for general tools).
Top: Gemini 2.5 Pro (73.3% reliable first drafts)
Run by an independent project, not by a vendor, but the site also sells benchmarking and some tool descriptions carry affiliate-tagged links; 30 tasks and one prompt per tool with default features, so a snapshot of 2025 versions that is over a year old; an LLM jury graded reliability with expert review of borderline cases, and usefulness and workflow scores are expert opinion; one tool is anonymized; the authors say their 2025 figures are not comparable with the current leaderboard pass rates; read from the live page 2026-10-23.
legalbenchmarks.ai/research/phase-2-research97 legal AI products in 18 categories (checked 2026-10-23), from AI receptionists that answer a firm’s phone to e-discovery and contract lifecycle tools. 11 have enough voices to be scored, 42 are named by a few voices, the rest have none yet. A tool with few voices is listed as such and never given a score. Missing a tool? Tell us; listing is free and nobody pays for a position.
| Tool | Vendor | In our record | What it does | |
|---|---|---|---|---|
| Legal AI assistant 12 | ||||
| Harvey agent | Harvey | scored · 749 voices | Legal AI platform for law firms and corporate legal teams: contract analysis, due diligence, compliance and litigation workflows, plus agents and a document vault. | site |
| CoCounsel agent | Thomson Reuters | scored · 411 voices | Thomson Reuters AI assistant for legal, tax and other professionals; grounded in Thomson Reuters content. Casetext's product line is folded into it. | site |
| Legora agent | Legora | scored · 90 voices | Collaborative legal AI workspace for review, research, drafting and advising, aimed at law firms and in-house teams. | site |
| agent | Paxton | scored · 12 voices | AI workspace for litigation teams that combines case files and legal research. | site |
| Casetext | Thomson Reuters (acquired 2023) | few voices · 117 mentions | Original CoCounsel maker; casetext.com now redirects to Thomson Reuters CoCounsel. | site |
| GC AI agent | GC AI | few voices · 26 mentions | AI platform for in-house legal teams covering drafting, review, playbooks and connected workflows. | site |
| August (formerly Vecflow / Oliver) agent | August | not yet in the record | Legal AI workspace for law firms (document review, drafting, invoicing). Formerly Vecflow; its Oliver assistant was in the Vals VLAIR Feb 2025 report. vecflow.com is now an unrelated site. | site |
| Eudia agent | Eudia | not yet in the record | AI platform for legal, procurement and government acquisition teams. | site |
| Flank agent | Flank | not yet in the record | AI agents for enterprise legal teams that handle repeat requests like NDAs, drafting, redlining and Q&A via email or web. | site |
| Jylo agent | Jylo | not yet in the record | Private AI workspace for professional firms with playbooks, agents and model controls. | site |
| Legalfly agent | Legalfly | not yet in the record | Legal AI for corporates: intake, contract review, drafting, due diligence and research. | site |
| Norm Ai agent | Norm Ai | not yet in the record | Compliance-oriented legal AI agents ('Agentic Law'). Reported USD 120M Series C in July 2026 (search coverage, not verified at source). | site |
| Research platform 12 | ||||
| Lexis+ AI agent | LexisNexis | scored · 294 voices | LexisNexis legal AI for research, drafting and analysis over LexisNexis content; the old lexis-plus-ai URL now redirects to the Lexis+ with Protégé page. | site |
| Westlaw AI | Thomson Reuters | scored · 268 voices | Thomson Reuters legal research tools with AI layers; Westlaw Advantage is the current AI research page, Westlaw Edge and Precision are older names in the same line. | site |
| vLex Vincent agent | vLex (part of Clio) | scored · 31 voices | AI legal research platform; the site calls vLex 'part of Clio'. Vincent is its AI assistant. | site |
 Midpage | Midpage | few voices · 18 mentions | AI legal research and drafting over US case law, statutes and regulations; participant in the Vals VLAIR legal research study (Oct 2025). | site |
 Descrybe | Descrybe | few voices · 6 mentions | Legal research platform and legal research engine that works in its own app and inside ChatGPT, Claude, Microsoft 365 Copilot and Perplexity. descrybe.ai says it moved to descrybe.com. | site |
| Alexi agent | Alexi | few voices · 3 mentions | AI platform for law firms covering legal research, drafting and analysis. Was a participant in the Vals VLAIR legal research study (Oct 2025). | site |
| DeepJudge agent | DeepJudge | few voices · 2 mentions | Enterprise search over a firm's own knowledge with a layer to build AI applications, workflows and agents. | site |
| Bloomberg Law AI | Bloomberg Law | not yet in the record | Legal research platform from Bloomberg with AI features. Homepage blocks automated fetches, so the product description is not independently re-read. | site |
| Counsel Stack | Counsel Stack | not yet in the record | Citation verification MCP and API for legal AI; top scorer in the Vals VLAIR legal research study (Oct 2025). | site |
| Lexroom | Lexroom | not yet in the record | Italian legal AI for research, document analysis and drafting with verified sources. Reported USD 50M round in May 2026 (search coverage, not verified at source). | site |
 OpenCase | OpenCase | not yet in the record | Legal AI that answers research questions with case law, statutes and regulations, with citations verified against primary sources. | site |
 Qura | Qura | not yet in the record | Legal database platform covering 1000+ sources including case law, legislation and regulations, connected to a firm's internal knowledge. Not the social media tool at qura.ai. | site |
| Contract drafting 4 | ||||
| agent | Spellbook (Rally Legal) | scored · 99 voices | Contract review and drafting assistant that runs inside Microsoft Word; the site says it uses GPT-5, Claude and other LLMs. | site |
| DraftWise agent | DraftWise | not yet in the record | Contract drafting, review and negotiation AI that draws on a firm's own precedent and guidance. | site |
| Henchman | Henchman (LexisNexis) | not yet in the record | Contract drafting tool, announced as acquired by LexisNexis in 2024 (trade press); its capabilities were slated for Lexis+ AI and Lexis Create. henchman.io no longer answers. | site |
| Wordsmith agent | Wordsmith | not yet in the record | AI platform for in-house legal teams: drafting, reviewing and researching contracts. Reported USD 70M Series B in June 2026 (search coverage, not verified at source). | site |
| Drafting 2 | ||||
 Law Insider | Law Insider | few voices · 7 mentions | AI contract drafting, review and redlining backed by a large public contract database, with a Word add-in. | site |
| Laine | Laine Neural Network SA | not yet in the record | Legal AI platform that drafts 50+ document types, edits with tracked changes, reviews documents and applies team playbooks, for law firms and in-house teams. Read through a fetch tool because direct requests returned 403. | site |
| Contract review / CLM 16 | ||||
| Gavel (acquired by Relativity) | few voices · 26 mentions | Reviews, redlines and drafts commercial contracts natively in Word and online, and turns legal intake into documents; the homepage says Gavel has been acquired by Relativity. | site | |
| Luminance agent | Luminance | few voices · 9 voices | Contract intelligence platform for reviewing, negotiating and managing contracts across an enterprise. | site |
| LegalOn agent | LegalOn Technologies | few voices · 7 mentions | Legal AI for contract review with attorney-built playbooks; site claims 9,000+ teams. | site |
| Robin AI | Robin AI | few voices · 5 mentions | AI contract review product. Secondary sources report the company wound down in late 2025 and sold its assets (managed services to Scissero; engineers reportedly to Microsoft). The homepage still serves marketing copy. Status unverified at primary source. | site |
| Chamelio agent | Chamelio | few voices · 2 mentions | In-house legal platform for contract management, review, intake and an agent layer for routing and redlining. | site |
 Dioptra | Dioptra (acquired by Icertis) | few voices · 2 mentions | Generates redlines in Microsoft Word, builds custom playbooks and flags non-market risks; the site banner says Icertis acquires Dioptra. | site |
 Agiloft | Agiloft | few voices · 1 mention | AI contract lifecycle management to draft, redline, negotiate and analyze contracts. | site |
| Definely agent | Definely | few voices · 1 mention | Structured review tools to help lawyers read, analyse and proof complex contracts. | site |
| Evisort (Workday Contract Intelligence) | Evisort (Workday) | few voices · 1 mention | Workday Contract Intelligence, powered by Evisort AI, is a contract repository with an Ask AI feature, term extraction and custom AI models across all contracts. | site |
| Ironclad agent | Ironclad | few voices · 1 mention | Contract lifecycle management platform with AI for contract creation and insights. | site |
 Lexion / Docusign IAM | Lexion (acquired by Docusign) | few voices · 1 mention | Docusign Intelligent Agreement Management is its agreement platform, with legal listed among its departments; lexion.ai carries a banner that Lexion joins forces with Docusign, and Docusign's blog says the Lexion acquisition is complete. | site |
| Checkbox | Checkbox | not yet in the record | AI legal front door for in-house teams to capture, triage, automate, manage and report on legal work across the tools they already use. | site |
 Ivo | Ivo | not yet in the record | AI contract review, redlining and intelligence for in-house teams; SOC 2 and ISO 27001 claimed. | site |
| Juro agent | Juro | not yet in the record | Contract management platform that runs contracts end to end with AI automation. | site |
| Lawgeex | Lawgeex | not yet in the record | Automated contract review and redlining against set policies. One aggregator (CB Insights) lists it as acquired by LegalSifter in 2023; not confirmed at a primary source. | site |
 Superlegal | Superlegal | not yet in the record | AI-plus-attorney contract review and redlining service for builders. | site |
| Lease / real estate 4 | ||||
 Acrebase | Acrebase | not yet in the record | Models how a commercial lease negotiation moves, covering what the other side will accept, fight or concede on each deal point, with six tools on one model and deep coverage of Texas retail leases. | site |
| LeasePilot | LeasePilot | not yet in the record | Commercial lease drafting platform that produces first drafts in minutes; the site claims 33,000+ commercial leases across 50 states. | site |
 Orbital | Orbital (formerly Orbital Witness) | not yet in the record | AI platform for commercial real estate legal work, from asset intelligence to abstracting, drafting and negotiating leases and amendments. orbitalwitness.com redirects to orbital.tech. | site |
 Prophia | Prophia | not yet in the record | AI lease abstraction and portfolio insights for commercial real estate. | site |
| Due diligence / search 3 | ||||
 Hebbia | Hebbia | few voices · 1 mention | AI platform for document-heavy analysis; its homepage now leads with finance, and names law firms among its users. | site |
| Kira (Litera) agent | Litera | few voices · 1 mention | Contract review and due diligence extraction, now sold as Litera Review; the old litera.com/products/kira URL redirects there. | site |
| Trellis | Trellis | not yet in the record | State court litigation research and analytics platform. Homepage is behind a Cloudflare challenge, so this line is from general knowledge, not a fresh read. | site |
| Deal / M&A analytics 1 | ||||
 Centari | Centari | not yet in the record | Converts deal documents into structured data for law firms and investment funds. The page carries a banner saying Centari has evolved and pointing to a new brand. | site |
| E-discovery 8 | ||||
 Everlaw | Everlaw | few voices · 6 mentions | Cloud e-discovery platform with AI features for litigation and investigations. | site |
| Relativity aiR agent | Relativity | few voices · 4 mentions | Generative AI document review inside the Relativity e-discovery platform; the old aiR URL redirects to the platform page. | site |
 Logikcull | Logikcull (listed as a Reveal product on revealdata.com) | few voices · 1 mention | Cloud-based eDiscovery and legal holds software for disputes, subpoenas and investigations. | site |
| DISCO agent | DISCO | not yet in the record | Cloud e-discovery software and services with AI features. | site |
 Exterro | Exterro | not yet in the record | Data risk management platform combining eDiscovery, digital forensics, privacy and data governance. | site |
 Lighthouse | Lighthouse | not yet in the record | AI-powered eDiscovery and legal intelligence with expert legal services. | site |
| Nuix | Nuix | not yet in the record | Investigative analytics and intelligence software covering data privacy, fraud and investigations, and legal eDiscovery. | site |
 Reveal | Reveal | not yet in the record | AI-driven eDiscovery software for corporate legal, law firms, LSPs, government and education; its product list also includes Logikcull, Brainspace, Reveal Hold and Trial Director. | site |
| Litigation / PI 8 | ||||
| Supio agent | Supio | few voices · 9 mentions | Legal AI for plaintiff and personal injury firms, from intake through demands and litigation drafting. | site |
| Clearbrief | Clearbrief | few voices · 8 mentions | Word add-in that finds supporting evidence for sentences in a brief and builds tables of authorities. | site |
| EvenUp agent | EvenUp | few voices · 6 mentions | AI platform for personal injury law firms, covering demand packages and case automation. | site |
 Briefpoint | Briefpoint | few voices · 3 mentions | Drafts discovery documents (requests for production, admissions, interrogatories) with AI. | site |
| Eve agent | Eve | few voices · 2 mentions | Legal AI software for plaintiff firms; the site claims 1,200+ firms. | site |
| CaseMark (Para) agent | CaseMark | few voices · 1 mention | AI colleague for legal teams (medical chronologies to motions) driven by email or Teams; the site says 'Formerly CaseMark Operator'. casemark.ai redirects to casemark.com. | site |
 StrongSuit (formerly Callidus) | StrongSuit | few voices · 1 mention | Legal AI for litigators: case law research, document review and drafting. Renamed from Callidus Legal AI in Nov 2025 per LawSites; callidusai.com redirects to strongsuit.com. | site |
 Darrow | Darrow | not yet in the record | AI legal intelligence that surfaces violations and connects plaintiffs' attorneys with high-value cases. | site |
| Litigation / analytics 2 | ||||
| Lex Machina | Lex Machina (LexisNexis) | few voices · 2 mentions | Legal analytics on judges, courts, counsel and parties to support strategy and case assessment, now empowered by LexisNexis Protege. | site |
 DepoIQ (InFactIQ) | InFactIQ (depoiq.com redirects there) | not yet in the record | Reads depositions, medical records, claims files, bills and testimony to flag contradictions and gaps for claims and litigation teams, citing each finding to its source. The page does not mention the DepoIQ name; depoiq.com redirects to infactiq.com. | site |
| Intake / AI receptionist 7 | ||||
 Lawmatics | Lawmatics | few voices · 2 mentions | Legal CRM and client intake platform for law firms; its Merlin Qualify AI scores leads so firms can sign the right clients faster. | site |
 Smith.ai | Smith.ai | few voices · 2 mentions | AI receptionists plus live human agents that answer law firm calls 24/7, qualify leads and book consultations, with details pushed to the firm's CRM. | site |
| LegalClerk.ai | LegalClerk | few voices · 1 mention | 24/7 AI answering service and virtual receptionist for law firms that answers, qualifies and captures every call. | site |
 HelloCounsel | HelloCounsel | not yet in the record | AI agents for personal injury, disability and mass tort firms covering reception, intake, case opening, medical records and client updates. | site |
 Intaker | Intaker | not yet in the record | AI client intake software for law firms with website chat, two-way texting, an AI texting agent and a Google LSA responder. | site |
 Lexidesk | Lexidesk | not yet in the record | AI intake for law firms across phone and website chat that answers instantly, qualifies every lead and delivers the results to the firm's CRM. | site |
 Meet Gabbi | Meet Gabbi | not yet in the record | Voice AI and texting intake for law firms that qualifies cases and captures client data in real time. | site |
| Practice management 6 | ||||
| Clio | Clio | scored · 78 voices | Practice management software with AI features (Clio Duo; Clio Work with vLex research content). Clio completed its USD 1B vLex acquisition in Nov 2025 per press coverage. | site |
 MyCase (8am IQ) | 8am (MyCase) | few voices · 11 mentions | Legal practice management software with 8am IQ AI tools for case assistance, document processing and legal writing built in. | site |
| Filevine AI (LOIS) agent | Filevine | few voices · 10 mentions | Legal work platform and practice management with its LOIS AI layer. | site |
 Smokeball | Smokeball | few voices · 8 mentions | Legal practice management software with automatic time tracking (AutoTime) and Archie AI, an assistant that drafts, reviews and answers questions inside the matter. | site |
 Actionstep Intelligence | Actionstep | few voices · 1 mention | Law firm management platform with AI features for time capture, client intake, filing and administration. | site |
| CosmoLex | CosmoLex | not yet in the record | Cloud legal practice management and accounting software that advertises AI intake forms and lead tracking. | site |
| Billing / time capture 4 | ||||
 Ajax | Ajax | few voices · 1 mention | AI timekeeper for lawyers that tracks work and writes custom, OCG-compliant time entries for the lawyer to review and release. | site |
| Billables AI | Billables AI | few voices · 1 mention | Automated time tracking for law firms that turns daily activity into ready-to-review time entries with client, matter and narrative. | site |
 Laurel | Laurel | not yet in the record | AI timekeeping platform that captures billable work across the tools a team uses and generates compliant time narratives. | site |
 PointOne | PointOne | not yet in the record | AI timekeeping and bill review for law firms that passively captures time and applies billing rules before bills go out. | site |
| Immigration 4 | ||||
| Docketwise (8am IQ) | 8am (Docketwise) | few voices · 1 mention | Immigration case management with forms automation and 8am IQ AI that captures client and case data straight into forms. | site |
 Visalaw AI | Visalaw AI | few voices · 1 mention | AI research, drafting and document analysis platform for immigration lawyers, with cited answers and a Research Agent in version 2.1. | site |
 CaseBlink | CaseBlink | not yet in the record | Immigration firm system of record where AI agents handle intakes, forms preparation, RFE responses, packet assembly, deadlines and time tracking. | site |
| eimmigration | Cerenade | not yet in the record | All-in-one immigration law software for cases, forms and client management, with an AI Workmates feature and AI Mode. | site |
| Tax research 1 | ||||
 Blue J | Blue J | not yet in the record | AI tax research platform that gives tax research answers in seconds; the site says it is trusted by 6,000+ firms. | site |
| General model (legal offering) 2 | ||||
| Claude | Anthropic | scored · 872 voices | Anthropic's legal solutions page says Claude helps legal teams and law firms with contract review, research, drafting and diligence. | site |
| Gemini | Google Cloud | scored · 134 voices | Google Cloud's page calls Gemini for legal enterprise AI purpose-built for modern legal practice; the press release lists it as Gemini Enterprise for Legal, initially in preview. | site |
| AI-first law firm (service) 1 | ||||
| Crosby agent | Crosby | not yet in the record | AI-first law firm that reviews commercial contracts using agents with attorneys reviewing the work. It is a law firm service, not software you license. | site |
Short answers with the date and the sample, written to be quoted.
On Oct 26, 2026, Claude: among tools with 60+ counted voices it has the highest floor of net sentiment (+37, 90% band +33 to +41, 872 voices). The answer depends on the task; the task leaders are listed at the top of this page.
Perplexity leads legal research by band floor (+88 net on 17 voices, Low confidence); it is tied with Clio, vLex Vincent, so the order is not settled.
Claude leads drafting by band floor (+57 net on 141 voices, Medium confidence); it is tied with Gemini, Spellbook, so the order is not settled.
Claude leads contract review by band floor (+46 net on 65 voices, Medium confidence); it is tied with Spellbook, Harvey, so the order is not settled.
The newest flagship of each lab, by release date: DeepSeek V4.1 Flash (DeepSeek, 2026-10-17); Gemini 4 Argon (Google, 2026-10-13); Muse Wonder Full (Meta, 2026-10-12); Mistral Large 4 (Mistral, 2026-10-06); GPT-6 Astra (OpenAI, 2026-09-03); Claude Fable 5.1 (Anthropic, 2026-09-01); Grok 4.7 (xAI, 2026-09 (month only)); Kimi K3 (Moonshot, date not published); Qwen3.8 Max (Alibaba, date not published); Command A+ (Cohere, date not published). On Vals Legal Research Bench (All-Pass accuracy), an outside test of agentic US legal research read 2026-10-23, the top score is 55.29% by Claude Fable 5.1, Claude Opus 5, Muse Spark 1.3. That is a test score, not what lawyers report; see Frontier models.
An agent takes a legal task and works through its steps (reads documents, runs a review, drafts, checks) instead of answering one prompt. 33 legal AI products ship agents today, including Harvey Agents (500+ prebuilt agents and Agent Builder), Legora aOS / Legora Agent, CoCounsel Legal (Deep Research, guided workflows, Westlaw Brief Builder, Drafting Agent), Protégé General AI (Orchestrator, Legal Research and Reflection agents), Agent Lumi, Vincent Agentic Mode (inside Clio Work), Communication Agents, Eve Agents (Eve 2.0). Each with its source.
Yes. 7 AI receptionist and intake products for law firms are in our market list: Smith.ai, Lexidesk, Meet Gabbi, LegalClerk.ai, Intaker, HelloCounsel, Lawmatics. They answer calls or chats, qualify the lead and book a consult; check each vendor page for what it does without a person. The list.
Lawyers are split to negative on it: net -13 on 2,467 voices (90% band -16 to -10); the complaint named most is hallucination.
Lawyers are split to negative on it: net -21 on 749 voices (90% band -26 to -15); the complaint named most is accuracy.
The preregistered study found Lexis+ AI and Thomson Reuters' Westlaw AI-Assisted Research and Ask Practical Law AI hallucinated 17% to 33% of the time per the arXiv abstract, fewer than GPT-4. The HAI post puts Lexis+ AI and Ask Practical Law AI at more than 17% and Westlaw AI-Assisted Research at more than 34%. (Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (Magesh et al.), published 2024-05-30 (arXiv v1; HAI news post dated 2024-05-23); that is an outside test, not our score.)
Each voice is a public post by someone using the tool, kept only if it states an experience, a verdict or a fact about it. Net is likes minus dislikes per 100 voices; tools are ordered by the floor of the 90% band. Nobody pays to be listed or for a position. Method · Independence.
