ROHAS

ROHAS Legal AI (India) Benchmark

10 questions test reasoning, authority accuracy, honesty about missing information, applied judgment and instruction fidelity for a total of 100 marks. Two serious failures are tracked separately as disqualifiers.

R Reasoning & Risk O Origin & Accuracy H Honesty about gaps A Applied context S Structure & Fidelity

How it works Results archive

Run configuration

Temperature0 where supported; provider default otherwise

Web & toolsDisabled

Version 3.1Provider-default reasoning · 4,000 answer tokens · no retry

Version 3.2Low reasoning · 6,000 answer tokens · one technical retry

Benchmark costAnswer calls only; judge excluded

Final gradingAdministrator reviewed

How to read these results

Performance is measured on a fixed Indian-law benchmark. Results reflect the particular model, provider, configuration, test date and administrator-reviewed grading used in that run. They do not establish a model's accuracy or suitability for every legal task. This is a research benchmark, not legal advice, and does not create a lawyer-client relationship. Law is stated as at the date disclosed in the benchmark. Model names and trademarks belong to their respective owners; ROHAS is not affiliated with or endorsed by any model provider.

ROHAS Legal AI (India) Benchmark Leaderboard

A model that triggers either explicit disqualifier cannot rank #1, however high its raw score.

Order by:

Anthropic: Claude Opus 5

anthropic/claude-opus-5 · v3.2

Winner

98%

R 100%

O 100%

H 90%

A 100%

S 100%

Answer cost $0.4864 Runtime 5m 51s

v3.2 · 28 August 2026

OpenAI: GPT-5.6 Sol

openai/gpt-5.6-sol · v3.1

97%

R 95%

O 100%

H 95%

A 100%

S 95%

Answer cost $0.1955 Runtime 6m 4s

v3.1 · 28 August 2026

OpenAI: GPT-5.6 Terra

openai/gpt-5.6-terra · v3.1

95%

R 100%

O 90%

H 95%

A 95%

S 95%

Answer cost $0.1778 Runtime 5m 22s

v3.1 · 28 August 2026

OpenAI: GPT-5.6 Luna

openai/gpt-5.6-luna · v3.1

94%

R 95%

O 95%

H 90%

A 90%

S 100%

Answer cost $0.0189 Runtime 5m 6s

v3.1 · 28 August 2026

Anthropic: Claude Sonnet 5

anthropic/claude-sonnet-5 · v3.1

92%

R 95%

O 95%

H 85%

A 90%

S 95%

Answer cost $0.1638 Runtime 5m 6s

v3.1 · 28 August 2026

MoonshotAI: Kimi K3

moonshotai/kimi-k3 · v3.2

91%

R 95%

O 95%

H 85%

A 85%

S 95%

Answer cost $0.1135 Runtime 2m 39s

v3.2 · 28 August 2026

Qwen: Qwen3.8 Max

qwen/qwen3.8-max · v3.2

89%

R 95%

O 85%

H 85%

A 80%

S 100%

Answer cost $0.0961 Runtime 6m 19s

v3.2 · 28 August 2026

DeepSeek: DeepSeek V4 Pro 0813

deepseek/deepseek-v4-pro-0813 · v3.2

87%

R 90%

O 90%

H 75%

A 85%

S 95%

Answer cost $0.1275 Runtime 13m 37s

v3.2 · 28 August 2026

SpaceXAI: Grok 4.6

x-ai/grok-4.6 · v3.2

86%

R 90%

O 90%

H 80%

A 80%

S 90%

Answer cost $0.0459 Runtime 3m 26s

v3.2 · 28 August 2026

#10

Z.ai: GLM 5.3

z-ai/glm-5.3 · v3.2

83%

R 85%

O 80%

H 80%

A 75%

S 95%

Answer cost $0.0306 Runtime 3m 18s

v3.2 · 28 August 2026

#11

Google: Gemini 3.7 Flash

google/gemini-3.7-flash · v3.2

80%

R 95%

O 60%

H 70%

A 80%

S 95%

Answer cost $0.0398 Runtime 2m 42s

v3.2 · 29 August 2026

#12

Perplexity: Sonar

perplexity/sonar · v3.2

76%

R 85%

O 75%

H 65%

A 60%

S 95%

Answer cost $0.0552 Runtime 1m 55s

v3.2 · 29 August 2026

#13

Anthropic: Claude Haiku 4.5

anthropic/claude-haiku-4.5 · v3.1

72%

R 85%

O 70%

H 70%

A 45%

S 90%

Answer cost $0.0309 Runtime 3m 11s

v3.1 · 28 August 2026

#14

Amazon: Nova Premier 1.0

amazon/nova-premier-v1 · v3.2

61%

R 85%

O 80%

H 55%

A 20%

S 65%

Answer cost $0.0420 Runtime 1m 56s

v3.2 · 29 August 2026

Full data ▼

Rank	Model	Version	Score	R / O / H / A / S	Answer cost	Runtime	Run
#1	Anthropic: Claude Opus 5 anthropic/claude-opus-5	v3.2	98 / 100 (98%)	100% / 100% / 90% / 100% / 100%	$0.4864	5m 51s	2026-08-28 10:50:44
#2	OpenAI: GPT-5.6 Sol openai/gpt-5.6-sol	v3.1	97 / 100 (97%)	95% / 100% / 95% / 100% / 95%	$0.1955	6m 4s	2026-08-28 15:35:21
#3	OpenAI: GPT-5.6 Terra openai/gpt-5.6-terra	v3.1	95 / 100 (95%)	100% / 90% / 95% / 95% / 95%	$0.1778	5m 22s	2026-08-28 08:50:44
#4	OpenAI: GPT-5.6 Luna openai/gpt-5.6-luna	v3.1	94 / 100 (94%)	95% / 95% / 90% / 90% / 100%	$0.0189	5m 6s	2026-08-28 17:15:09
#5	Anthropic: Claude Sonnet 5 anthropic/claude-sonnet-5	v3.1	92 / 100 (92%)	95% / 95% / 85% / 90% / 95%	$0.1638	5m 6s	2026-08-28 16:44:37
#6	MoonshotAI: Kimi K3 moonshotai/kimi-k3	v3.2	91 / 100 (91%)	95% / 95% / 85% / 85% / 95%	$0.1135	2m 39s	2026-08-28 11:27:05
#7	Qwen: Qwen3.8 Max qwen/qwen3.8-max	v3.2	89 / 100 (89%)	95% / 85% / 85% / 80% / 100%	$0.0961	6m 19s	2026-08-28 11:36:23
#8	DeepSeek: DeepSeek V4 Pro 0813 deepseek/deepseek-v4-pro-0813	v3.2	87 / 100 (87%)	90% / 90% / 75% / 85% / 95%	$0.1275	13m 37s	2026-08-28 11:22:16
#9	SpaceXAI: Grok 4.6 x-ai/grok-4.6	v3.2	86 / 100 (86%)	90% / 90% / 80% / 80% / 90%	$0.0459	3m 26s	2026-08-28 11:02:48
#10	Z.ai: GLM 5.3 z-ai/glm-5.3	v3.2	83 / 100 (83%)	85% / 80% / 80% / 75% / 95%	$0.0306	3m 18s	2026-08-28 11:41:34
#11	Google: Gemini 3.7 Flash google/gemini-3.7-flash	v3.2	80 / 100 (80%)	95% / 60% / 70% / 80% / 95%	$0.0398	2m 42s	2026-08-29 03:48:55
#12	Perplexity: Sonar perplexity/sonar	v3.2	76 / 100 (76%)	85% / 75% / 65% / 60% / 95%	$0.0552	1m 55s	2026-08-29 03:43:30
#13	Anthropic: Claude Haiku 4.5 anthropic/claude-haiku-4.5	v3.1	72 / 100 (72%)	85% / 70% / 70% / 45% / 90%	$0.0309	3m 11s	2026-08-28 09:19:06
#14	Amazon: Nova Premier 1.0 amazon/nova-premier-v1	v3.2	61 / 100 (61%)	85% / 80% / 55% / 20% / 65%	$0.0420	1m 56s	2026-08-29 03:58:30