Vals LegalBench (LegalBench)

We show this table for reference; we do not rank on it.

Data verified October 13, 2026

Vals AI legal benchmark with issue, rule, conclusion, interpretation, and rhetoric task views.

Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.

LegalBench score on LegalBench — October 13, 2026

We mirror the published legalbench score view for LegalBench. Claude Fable 5 leads the public snapshot at 88.56%, followed by Claude Fable 5.1 (88.51%) and Gemini 4 Argon (88.30%). We do not use these results to rank models overall.

1Closed

Claude Fable 5

Anthropic

88.56%

Overall 80.78Context 1M+

2Closed

Claude Fable 5.1

Anthropic

88.51%

Overall 81.08Context 1M

3Closed

Gemini 4 Argon

Google

high reasoning

88.30%

Overall 82.28Context 1M

151 modelsKnowledgeCurrentDisplay onlyUpdated October 13, 2026

LegalBench score table (151 models)

LegalBench score results for LegalBench
Rank	Model / configuration	Score	Parameters (B)	Open / closed
1	Claude Fable 5Anthropic	88.56%	Not reported	Closed
2	Claude Fable 5.1Anthropic	88.51%	Not reported	Closed
3	Gemini 4 ArgonGooglehigh reasoning	88.30%	Not reported	Closed
4	Muse Wonder FullMetamax reasoning	87.42%	Not reported	Closed
5	Gemini 3.1 Pro PreviewGooglehigh reasoning	87.40%	Not reported	Unknown
6	Gemini 3.7 FlashGooglehigh reasoning	87.26%	Not reported	Closed
7	Gemini 3 Pro PreviewGooglehigh reasoning	87.03%	Not reported	Unknown
8	Gemini 3.8 FlashGooglehigh reasoning	86.99%	Not reported	Closed
9	Claude Opus 5Anthropic	86.97%	Not reported	Closed
10	GPT-5.6 SolOpenAImax reasoning	86.97%	Not reported	Closed
11	Gemini 3 Flash PreviewGooglehigh reasoning	86.86%	Not reported	Unknown
12	Gemini 3.6 FlashGooglehigh reasoning	86.70%	Not reported	Closed
13	GPT-5.5OpenAIxhigh reasoning	86.52%	Not reported	Closed
14	Grok 4.6xAIhigh reasoning	86.31%	Not reported	Closed
15	GLM LegalZhipu AImax reasoning	86.23%	Not reported	Unknown
16	Kimi K3Moonshot AImax reasoning	86.21%	Not reported	Open
17	GPT-5.4OpenAIxhigh reasoning	86.04%	Not reported	Closed
18	GPT-5OpenAIhigh reasoning	86.02%	Not reported	Unknown
19	Grok 4.5xAIhigh reasoning	85.97%	Not reported	Closed
20	GPT-5.1OpenAIhigh reasoning	85.68%	Not reported	Closed
21	MiniMax M3MiniMax	85.42%	Not reported	Open
22	Claude Opus 4.6 (Adaptive)Anthropic	85.30%	Not reported	Closed
23	Muse Spark 1.2Metaxhigh reasoning	85.26%	Not reported	Closed
24	Claude Opus 4.7Anthropic	85.25%	Not reported	Closed
25	GPT-5.6 TerraOpenAImax reasoning	85.11%	Not reported	Closed
26	Qwen3.5 Plus ThinkingAlibaba	85.10%	Not reported	Unknown
27	Muse Spark 1.1Metaxhigh reasoning	84.98%	Not reported	Closed
28	Qwen3.7 MaxAlibaba	84.91%	Not reported	Closed
29	GLM-5.3Z.AImax reasoning	84.84%	Not reported	Open
30	Kimi K2.6Moonshot AI	84.74%	Not reported	Open
31	Claude Opus 4.5 ThinkingAnthropic	84.60%	Not reported	Closed
32	Grok 4.3xAI	84.46%	Not reported	Closed
33	GLM-5.1Z.AI	84.39%	Not reported	Open
34	Grok 4.7xAIxhigh reasoning	84.39%	Not reported	Closed
35	Gemini 2.5 Pro Exp 03 25Google	84.32%	Not reported	Unknown
36	Qwen3.5 FlashAlibaba	84.28%	Not reported	Closed
37	Qwen3.6 PlusAlibaba	84.23%	Not reported	Closed
38	Muse SparkMeta	84.22%	Not reported	Closed
39	Claude Sonnet 4.5 ThinkingAnthropic	84.08%	Not reported	Closed
40	Gemini 3.5 Flash-LiteGooglehigh reasoning	84.07%	Not reported	Closed
41	GLM-5.2Z.AI	84.07%	Not reported	Open
42	GLM 5 ThinkingZhipu AI	84.06%	Not reported	Unknown
43	GPT-5.6 LunaOpenAImax reasoning	84.03%	Not reported	Closed
44	MiniMax M2.7MiniMax	83.98%	Not reported	Open
45	GLM-5.3-FlashZ.AImax reasoning	83.93%	Not reported	Open
46	Claude Sonnet 5Anthropic	83.92%	Not reported	Closed
47	Gemini 2.5 Flash Preview 04 17Google	83.80%	Not reported	Unknown
48	Gemini 3.1 Flash Lite PreviewGooglehigh reasoning	83.76%	Not reported	Unknown
49	o3OpenAIhigh reasoning	83.76%	Not reported	Closed
50	Hy4 previewTencent	83.76%	Not reported	Open
51	Qwen3.8 MaxAlibaba	83.61%	Not reported	Open
52	Gemini 3.5 FlashGooglehigh reasoning	83.60%	Not reported	Closed
53	Claude Opus 4.8Anthropic	83.57%	Not reported	Closed
54	Claude Opus 4.1Anthropic	83.46%	Not reported	Unknown
55	GLM-4.7Z.AI	83.36%	Not reported	Open
56	DeepSeek V4.1 Flash 1017DeepSeekhigh reasoning	83.28%	Not reported	Open
57	Grok 4 0709xAI	83.19%	Not reported	Unknown
58	Grok 3 Mini Fast High ReasoningxAIhigh reasoning	83.14%	Not reported	Unknown
59	GPT-4.1OpenAIhigh reasoning	83.10%	Not reported	Closed
60	Claude Opus 4Anthropic	83.07%	Not reported	Unknown
61	Mercury 2.5Inceptionhigh reasoning	83.06%	Not reported	Closed
62	Inkling-SmallThinking Machines Lab0.99 reasoning	82.99%	Not reported	Open
63	Claude Sonnet 4Anthropic	82.95%	Not reported	Unknown
64	InklingThinking Machines Lab0.99 reasoning	82.95%	Not reported	Open
65	Claude Opus 4.5Anthropic	82.84%	Not reported	Closed
66	GPT-5.2OpenAIxhigh reasoning	82.76%	Not reported	Closed
67	Gemini 2.5 Flash Preview 09 2025 ThinkingGoogle	82.63%	Not reported	Unknown
68	Grok 3xAI	82.59%	Not reported	Unknown
69	Claude 3.7 Sonnet ThinkingAnthropic	82.52%	Not reported	Closed
70	Gemini 2.5 Flash Preview 09 2025Google	82.49%	Not reported	Unknown
71	Grok 4.1 Fast (Reasoning)xAI	82.45%	Not reported	Closed
72	Qwen3.8-27BAlibabaxhigh reasoning	82.43%	Not reported	Open
73	DeepSeek V4 Pro 0813DeepSeekmax reasoning	82.36%	Not reported	Open
74	GPT-4oOpenAIhigh reasoning	82.21%	Not reported	Closed
75	Claude Sonnet 4.6Anthropic	82.12%	Not reported	Closed
76	Nemotron 3 Ultra 550b A55bNvidia	82.07%	Not reported	Unknown
77	Claude Sonnet 4 20250514 ThinkingAnthropic	82.06%	Not reported	Unknown
78	Gemini 2.5 Flash Lite Preview 09 2025 ThinkingGoogle	82.04%	Not reported	Unknown
79	Grok 3 Mini Fast Low ReasoningxAIlow reasoning	81.92%	Not reported	Unknown
80	Qwen3 MaxAlibaba	81.86%	Not reported	Closed
81	GPT-5 miniOpenAIhigh reasoning	81.77%	Not reported	Closed
82	Moonshotai Kimi K2 InstructTogether AI	81.45%	Not reported	Unknown
83	Claude Haiku 4.5 ThinkingAnthropic	81.24%	Not reported	Closed
84	DeepSeek V3DeepSeek	80.76%	Not reported	Open
85	Grok 4 Fast (Reasoning)xAI	80.60%	Not reported	Closed
86	GPT-4 TurboOpenAIhigh reasoning	80.46%	Not reported	Closed
87	o1OpenAIhigh reasoning	80.39%	Not reported	Closed
88	Qwen3 Max PreviewAlibaba	80.33%	Not reported	Unknown
89	DeepSeek V4 Pro 0813DeepSeekmax reasoning	80.32%	Not reported	Open
90	Kimi K2 ThinkingMoonshot AI	80.20%	Not reported	Unknown
91	Qwen3 235B-A22BAlibaba	80.18%	Not reported	Open
92	GPT-4oOpenAIhigh reasoning	80.12%	Not reported	Closed
93	Claude 3.7 SonnetAnthropic	80.00%	Not reported	Closed
94	MiniMax M2.5MiniMax	79.96%	Not reported	Closed
95	Command A 03 2025Cohere	79.70%	Not reported	Unknown
96	Ling 3.0 Flash 2607Ant	79.69%	Not reported	Unknown
97	GLM-4.6Z.AI	79.61%	Not reported	Open
98	Qwen Qwen2.5 72B Instruct TurboTogether AI	79.40%	Not reported	Unknown
99	o4-miniOpenAIhigh reasoning	79.19%	Not reported	Closed
100	Mistral Large 2512Mistral AI	79.14%	Not reported	Unknown
101	Grok 4.1 Fast Non ReasoningxAI	79.11%	Not reported	Unknown
102	Gemini 2.5 Flash Lite Preview 09 2025Google	79.01%	Not reported	Unknown
103	MiMo-V2.5Xiaomi	78.89%	Not reported	Closed
104	Grok 4 Fast Non ReasoningxAI	78.40%	Not reported	Unknown
105	Gemini 2.0 Flash 001Google	78.36%	Not reported	Unknown
106	GPT-4.1 miniOpenAIhigh reasoning	78.04%	Not reported	Closed
107	GPT-5.4 nanoOpenAIhigh reasoning	77.92%	Not reported	Closed
108	Llama4 Maverick Instruct BasicFireworks AI	77.81%	Not reported	Unknown
109	Grok 4.20 0309 ReasoningxAI	77.74%	Not reported	Unknown
110	DeepSeek V3 0324DeepSeek	77.73%	Not reported	Open
111	DeepSeek V4 Flash 0731DeepSeekhigh reasoning	77.71%	Not reported	Open
112	Meta Llama Llama 3.3 70B Instruct TurboTogether AI	77.18%	Not reported	Unknown
113	MiMo-V2.5-ProXiaomi	77.13%	Not reported	Closed
114	DeepSeek V3p2 ThinkingFireworks AI	76.08%	Not reported	Unknown
115	GPT-OSS 120BOpenAI	75.94%	Not reported	Open
116	GLM-4.5Z.AI	75.63%	Not reported	Closed
117	MiniMax M2.1MiniMax	75.45%	Not reported	Unknown
118	Laguna M.1Poolside	75.14%	Not reported	Closed
119	Jamba 1.5 LargeAI21 Labs	74.16%	Not reported	Unknown
120	Meta Llama Llama 4 Scout 17B 16E InstructTogether AI	72.04%	Not reported	Unknown
121	Jamba Large 1.6AI21 Labs	71.96%	Not reported	Unknown
122	o3-miniOpenAIhigh reasoning	71.54%	Not reported	Closed
123	Laguna XS.2Poolside	71.03%	Not reported	Open
124	GPT-OSS 20BOpenAI	70.85%	Not reported	Open
125	Claude 3.5 HaikuAnthropic	70.33%	Not reported	Closed
126	Gemini 1.0 Pro 002Google	70.25%	Not reported	Unknown
127	Jamba Mini 1.6AI21 Labs	69.73%	Not reported	Unknown
128	Qwen Qwen2.5 7B Instruct TurboTogether AI	69.56%	Not reported	Unknown
129	Mistral Small 2503Mistral AI	69.16%	Not reported	Unknown
130	Gemini 1.5 Pro 002Google	69.08%	Not reported	Unknown
131	Ling 3.0 Flash Af Rc3Ant	68.74%	Not reported	Unknown
132	Command R PlusCohere	68.29%	Not reported	Unknown
133	Google Gemma 2 9b ItTogether AI	68.14%	Not reported	Unknown
134	Google Gemma 2 27b ItTogether AI	67.98%	Not reported	Unknown
135	DeepSeek-R1DeepSeek	67.32%	Not reported	Open
136	Jamba 1.5 MiniAI21 Labs	66.62%	Not reported	Unknown
137	DeepSeek V3p2Fireworks AI	65.96%	Not reported	Unknown
138	GPT-3.5 TurboOpenAIhigh reasoning	64.37%	Not reported	Unknown
139	Gemini 1.5 Flash 002Google	63.24%	Not reported	Unknown
140	Meta Llama Llama 2 70b HfTogether AI	62.43%	Not reported	Unknown
141	Mistral Medium 2505Mistral AI	61.98%	Not reported	Unknown
142	Command A Plus 05 2026Cohere	61.42%	Not reported	Unknown
143	GPT-4.1 nanoOpenAIhigh reasoning	61.06%	Not reported	Closed
144	Mistralai Mixtral 8x7B V0.1Together AI	55.84%	Not reported	Unknown
145	Magistral Medium 2509Mistral AI	54.77%	Not reported	Unknown
146	Mistralai Mistral 7B V0.1Together AI	53.77%	Not reported	Unknown
147	Togethercomputer Llama 2 13bTogether AI	53.70%	Not reported	Unknown
148	Togethercomputer Llama 2 7bTogether AI	51.87%	Not reported	Unknown
149	GPT-5 nanoOpenAIhigh reasoning	50.13%	Not reported	Closed
150	Magistral Small 2509Mistral AI	40.02%	Not reported	Unknown
151	Command RCohere	32.97%	Not reported	Unknown

How LegalBench is shown here

BenchLM mirrors the public Vals AI LegalBench leaderboard captured from https://www.vals.ai/benchmarks/legal_bench and updated by Vals on October 13, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

LegalBench is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.

Snapshot

151 Vals rows6 task viewspublic datasetTasks: Overall, Issue Tasks, Rule Tasks, Conclusion Tasks, Interpretation TasksDisplay only

LegalBench on Vals AI Vals methodology Vals home

The published LegalBench snapshot places Claude Fable 5 first at 88.56%. The third row is 0.26 points behind. The broader top-10 range is 1.60 points, so many of the published results sit in a relatively narrow band.

151 models have been evaluated on LegalBench. The benchmark falls in the Knowledge category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. LegalBench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About LegalBench

Year

2026

Tasks

Legal reasoning task views

Format

Accuracy score

Difficulty

Professional legal reasoning

BenchLM mirrors Vals LegalBench as display-only legal-domain evidence and does not use it in weighted rankings.

LegalBench Public benchmark source

Freshness and provenance

Version

LegalBench 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does LegalBench measure?

Vals AI legal benchmark with issue, rule, conclusion, interpretation, and rhetoric task views.

Which model leads the published LegalBench snapshot?

Claude Fable 5 currently leads the published LegalBench snapshot with 88.56% legalbench score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on LegalBench?

The October 13, 2026 snapshot contains 151 AI models.

Last updated: October 13, 2026 · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 6,100+ subscribers.

One email each week. Unsubscribe anytime.