Benchmark Almanac

SG-LegalClarify

Kimi K3 77.8% leads since November 2026
Not in model cards

Models answer Singapore legal questions and ask for missing facts. Three model judges grade each conversation against rubrics written by lawyers.

Made by
Ziwei Gu and colleagues
Released
November 2026
Kind
Law
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
All public
Measures
Capability
Subject
Law
Language
English
In and out
Text → Text
Interaction
Multi-turn
Refresh
One fixed set
Run by
Anyone
In views
Law and finance
Strongest model tested
Claude Fable 5 #12 on the capability index
Last model added
July 2026
Scores refreshed
November 20th, 2026
19
Entries in its paper, 16 models we track

Kimi K3 leads with 77.8% among models released by July 2026 Kimi K3 leads with 77.8% among models released by July 2026

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.

The newest model with a score here came out in July 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Kimi K3 Web enabled 77.8%
  2. 2 DeepSeek-V4-Flash Web enabled 74%
  3. 3 Claude Fable 5 Web enabled 73.3%
  4. 4 DeepSeek-V4-Pro Web enabled 73.1%
  5. 5 Claude Opus 4.8 Web enabled 72.9%
  6. 6 Gemini 3.5 Flash Web enabled 71.6%
  7. 7 GPT-5.6 Sol Web enabled 70.5%
  8. 8 Claude Sonnet 5 Web enabled 70.1%
  9. 9 Kimi K2.6 Web enabled 68.9%
  10. 10 Qwen3.7-Plus Web enabled 67.9%
  11. 11 GPT-5.6 Terra Web enabled 67.1%
  12. 12 GPT-5.6 Luna Web enabled 66.7%
  13. 13 Gemini 3.1 Pro Web enabled 59.9%
  14. 14 Qwen3.6 27B Web enabled 58.9%
  15. 15 Mistral 3 Large Web enabled 57.2%
  16. 16 Qwen-SEA-LION v4.5 27B Web enabled 54.7%
  17. 17 Gemma 3 27B Web enabled 37.7%
  18. 18 Llama 4 Scout Web enabled 37.5%
  19. 19 Gemma-SEA-LION v4 27B Web enabled 37.3%

Source: The SG-LegalClarify paper

#EntryScoreSettingDateSource
1 Kimi K3 77.8% Web enabled Jul 2026 Official leaderboard
2 DeepSeek-V4-Flash 74% Web enabled Apr 2026 Official leaderboard
3 Claude Fable 5 73.3% Web enabled Jun 2026 Official leaderboard
4 DeepSeek-V4-Pro 73.1% Web enabled Apr 2026 Official leaderboard
5 Claude Opus 4.8 72.9% Web enabled May 2026 Official leaderboard
6 Gemini 3.5 Flash 71.6% Web enabled May 2026 Official leaderboard
7 GPT-5.6 Sol 70.5% Web enabled Jul 2026 Official leaderboard
8 Claude Sonnet 5 70.1% Web enabled Jun 2026 Official leaderboard
9 Kimi K2.6 68.9% Web enabled Apr 2026 Official leaderboard
10 Qwen3.7-Plus 67.9% Web enabled Jun 2026 Official leaderboard
11 GPT-5.6 Terra 67.1% Web enabled Jul 2026 Official leaderboard
12 GPT-5.6 Luna 66.7% Web enabled Jul 2026 Official leaderboard
13 Gemini 3.1 Pro 59.9% Web enabled Feb 2026 Official leaderboard
14 Qwen3.6 27B 58.9% Web enabled Apr 2026 Official leaderboard
15 Mistral 3 Large 57.2% Web enabled Official leaderboard
16 Qwen-SEA-LION v4.5 27B 54.7% Web enabled Official leaderboard
17 Gemma 3 27B 37.7% Web enabled Mar 2025 Official leaderboard
18 Llama 4 Scout 37.5% Web enabled Apr 2025 Official leaderboard
19 Gemma-SEA-LION v4 27B 37.3% Web enabled Official leaderboard
#ModelScoreSettingDateSource
1 Kimi K3 77.8% Web enabled Jul 2026 Official leaderboard
2 DeepSeek-V4-Flash 74% Web enabled Apr 2026 Official leaderboard
3 Claude Fable 5 73.3% Web enabled Jun 2026 Official leaderboard
4 DeepSeek-V4-Pro 73.1% Web enabled Apr 2026 Official leaderboard
5 Claude Opus 4.8 72.9% Web enabled May 2026 Official leaderboard
6 Gemini 3.5 Flash 71.6% Web enabled May 2026 Official leaderboard
7 GPT-5.6 Sol 70.5% Web enabled Jul 2026 Official leaderboard
8 Claude Sonnet 5 70.1% Web enabled Jun 2026 Official leaderboard
9 Kimi K2.6 68.9% Web enabled Apr 2026 Official leaderboard
10 Qwen3.7-Plus 67.9% Web enabled Jun 2026 Official leaderboard
11 GPT-5.6 Terra 67.1% Web enabled Jul 2026 Official leaderboard
12 GPT-5.6 Luna 66.7% Web enabled Jul 2026 Official leaderboard
13 Gemini 3.1 Pro 59.9% Web enabled Feb 2026 Official leaderboard
14 Qwen3.6 27B 58.9% Web enabled Apr 2026 Official leaderboard
15 Gemma 3 27B 37.7% Web enabled Mar 2025 Official leaderboard
16 Llama 4 Scout 37.5% Web enabled Apr 2025 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Kimi K3
Closed book Web enabled
77.8%
DeepSeek-V4-Flash
Closed book Web enabled
74%
Claude Fable 5
Closed book Web enabled
73.3%
DeepSeek-V4-Pro
Closed book Web enabled
73.1%
Claude Opus 4.8
Closed book Web enabled
72.9%
Gemini 3.5 Flash
Closed book Web enabled
71.6%
GPT-5.6 Sol
Closed book Web enabled
70.5%
Claude Sonnet 5
Closed book Web enabled
70.1%
Kimi K2.6
Closed book Web enabled
68.9%
Qwen3.7-Plus
Closed book Web enabled
67.9%
GPT-5.6 Terra
Closed book Web enabled
67.1%
GPT-5.6 Luna
Closed book Web enabled
66.7%

Source: The SG-LegalClarify paper

ModelEffortHarnessScoreCost per taskOutput tokens
Kimi K3 Web enabled 77.8%
DeepSeek-V4-Flash Web enabled 74%
Claude Fable 5 Web enabled 73.3%
DeepSeek-V4-Pro Web enabled 73.1%
Claude Opus 4.8 Web enabled 72.9%
Gemini 3.5 Flash Web enabled 71.6%
GPT-5.6 Sol Web enabled 70.5%
Claude Sonnet 5 Web enabled 70.1%
Claude Fable 5 Closed book 69.8%
Kimi K2.6 Web enabled 68.9%
Qwen3.7-Plus Web enabled 67.9%
GPT-5.6 Terra Web enabled 67.1%
GPT-5.6 Luna Web enabled 66.7%
Claude Opus 4.8 Closed book 66.6%
Kimi K3 Closed book 65.9%
GPT-5.6 Sol Closed book 64.6%
Gemini 3.5 Flash Closed book 62.5%
Claude Sonnet 5 Closed book 61.2%
Gemini 3.1 Pro Web enabled 59.9%
GPT-5.6 Terra Closed book 59.8%
Qwen3.6 27B Web enabled 58.9%
GPT-5.6 Luna Closed book 58.4%
Gemini 3.1 Pro Closed book 57.2%
Kimi K2.6 Closed book 54.2%
DeepSeek-V4-Pro Closed book 52.4%
DeepSeek-V4-Flash Closed book 50.1%
Qwen3.7-Plus Closed book 49.9%
Qwen3.6 27B Closed book 47%
Gemma 3 27B Web enabled 37.7%
Llama 4 Scout Web enabled 37.5%
Llama 4 Scout Closed book 36%
Gemma 3 27B Closed book 35%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Kimi K3
Closed book Web enabled
77.8%
DeepSeek-V4-Flash
Closed book Web enabled
74%
Claude Fable 5
Closed book Web enabled
73.3%
DeepSeek-V4-Pro
Closed book Web enabled
73.1%
Claude Opus 4.8
Closed book Web enabled
72.9%
Gemini 3.5 Flash
Closed book Web enabled
71.6%
GPT-5.6 Sol
Closed book Web enabled
70.5%
Claude Sonnet 5
Closed book Web enabled
70.1%
Kimi K2.6
Closed book Web enabled
68.9%
Qwen3.7-Plus
Closed book Web enabled
67.9%
GPT-5.6 Terra
Closed book Web enabled
67.1%
GPT-5.6 Luna
Closed book Web enabled
66.7%

Source: The SG-LegalClarify paper

ModelEffortHarnessScoreCost per taskOutput tokens
Kimi K3 Web enabled 77.8%
DeepSeek-V4-Flash Web enabled 74%
Claude Fable 5 Web enabled 73.3%
DeepSeek-V4-Pro Web enabled 73.1%
Claude Opus 4.8 Web enabled 72.9%
Gemini 3.5 Flash Web enabled 71.6%
GPT-5.6 Sol Web enabled 70.5%
Claude Sonnet 5 Web enabled 70.1%
Claude Fable 5 Closed book 69.8%
Kimi K2.6 Web enabled 68.9%
Qwen3.7-Plus Web enabled 67.9%
GPT-5.6 Terra Web enabled 67.1%
GPT-5.6 Luna Web enabled 66.7%
Claude Opus 4.8 Closed book 66.6%
Kimi K3 Closed book 65.9%
GPT-5.6 Sol Closed book 64.6%
Gemini 3.5 Flash Closed book 62.5%
Claude Sonnet 5 Closed book 61.2%
Gemini 3.1 Pro Web enabled 59.9%
GPT-5.6 Terra Closed book 59.8%
Qwen3.6 27B Web enabled 58.9%
GPT-5.6 Luna Closed book 58.4%
Gemini 3.1 Pro Closed book 57.2%
Kimi K2.6 Closed book 54.2%
DeepSeek-V4-Pro Closed book 52.4%
DeepSeek-V4-Flash Closed book 50.1%
Qwen3.7-Plus Closed book 49.9%
Qwen3.6 27B Closed book 47%
Gemma 3 27B Web enabled 37.7%
Llama 4 Scout Web enabled 37.5%
Llama 4 Scout Closed book 36%
Gemma 3 27B Closed book 35%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Moonshot AI
Benchmark paper 1 leader
Kimi K3 77.8%

Source: The SG-LegalClarify paper

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Kimi K3 Kimi K2.6 Gemini 3.1 Pro Gemma 3 27B 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Kimi K3 Kimi K2.6

Source: Official SG-LegalClarify leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Law, finance and medicine

All 637 in Law, finance and medicine
  • MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy. Qwen 3.8 Max 80.4%
  • HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
  • HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
  • FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy. Ling-3.1-flash 81.7%
  • KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. o1-preview 92.7%
  • Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (19 entries).
  • Pass Models scored 16 models we track, plus 3 other entries on its leaderboard.
  • Pass Newest score The newest is from July 2026, for Kimi K3.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude Fable 5 (#12).
  • Pass Headroom The best is 77.8%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 32 rows for models we track and 3 other entries at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check