Benchmark Almanac

Immigration Bench

Claude Fable 5.1 75.4% leads since Oct 2026 Not in model cards

Models answer expert-written US immigration law questions from seven personas, such as F-1 students, employers and DACA recipients. An AI grader checks each answer against an expert rubric.

Experts wrote 519 questions. Vals reports on a private 262-question test set and reserves the other 257 for labs to license. Score is the share of rubric items met.

Made by Vals AI + Talent Mobility Fund (Renaissance Philanthropy) Released Oct 2026 Law Tasks kept private Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
Expert Specialists with years of training, often over hours of work
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Open-ended Free-form text, documents, designs or media judged as a whole.
Leaderboard vals.ai
24
Models scored on its own leaderboard

Claude Fable 5.1 leads with 75.4%

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

24 models. Scroll sideways for the rest.

  1. 75.4 Claude Fable 5.1
  2. 72.1 Claude Opus 5.5
  3. 71.2 Muse Wonder Full
  4. 70.1 Qwen 3.8 Max
  5. 69.6 Muse Wonder
  6. 69.0 Claude Sonnet 5.5
  7. 68.3 Kimi K3
  8. 68.1 Gemini 4 Argon
  9. 68.0 Grok 4.7
  10. 67.8 GPT-6.1 Sol
  11. 67.7 GPT-6 Astra
  12. 67.5 MiMo-V2.6 Pro
  13. 66.7 GLM-5.3
  14. 66.5 Muse Spark 1.2
  15. 66.0 Grok 4.6
  16. 65.6 DeepSeek V4.1 Flash
  17. 65.0 Muse Spark 1.3
  18. 63.7 Mistral Large 4
  19. 63.4 Gemini 3.8 Flash
  20. 62.5 GPT-6 Luna
  21. 61.8 Claude Haiku 5.5
  22. 61.7 DeepSeek V4 Pro 0813
  23. 57.5 MiniMax-M3
  24. 55.7 Gemini 3.5 Flash-Lite
  1. 1 Claude Fable 5.1 75.4%
  2. 2 Claude Opus 5.5 72.1%
  3. 3 Muse Wonder Full 71.2%
  4. 4 Qwen 3.8 Max 70.1%
  5. 5 Muse Wonder 69.6%
  6. 6 Claude Sonnet 5.5 69%
  7. 7 Kimi K3 68.3%
  8. 8 Gemini 4 Argon 68.1%
  9. 9 Grok 4.7 68%
  10. 10 GPT-6.1 Sol 67.8%
  11. 11 GPT-6 Astra 67.7%
  12. 12 MiMo-V2.6 Pro 67.5%
  13. 13 GLM-5.3 66.7%
  14. 14 Muse Spark 1.2 66.5%
  15. 15 Grok 4.6 66%
  16. 16 DeepSeek V4.1 Flash 65.6%
  17. 17 Muse Spark 1.3 65%
  18. 18 Mistral Large 4 63.7%
  19. 19 Gemini 3.8 Flash 63.4%
  20. 20 GPT-6 Luna 62.5%
  21. 21 Claude Haiku 5.5 61.8%
  22. 22 DeepSeek V4 Pro 0813 61.7%
  23. 23 MiniMax-M3 57.5%
  24. 24 Gemini 3.5 Flash-Lite 55.7%
#ModelScoreSettingDateSource
1 Claude Fable 5.1 75.4% Sep 2026 Official leaderboard
2 Claude Opus 5.5 72.1% Sep 2026 Official leaderboard
3 Muse Wonder Full 71.2% Oct 2026 Official leaderboard
4 Qwen 3.8 Max 70.1% Aug 2026 Official leaderboard
5 Muse Wonder 69.6% Oct 2026 Official leaderboard
6 Claude Sonnet 5.5 69% Sep 2026 Official leaderboard
7 Kimi K3 68.3% Jul 2026 Official leaderboard
8 Gemini 4 Argon 68.1% Sep 2026 Official leaderboard
9 Grok 4.7 68% Sep 2026 Official leaderboard
10 GPT-6.1 Sol 67.8% Sep 2026 Official leaderboard
11 GPT-6 Astra 67.7% Sep 2026 Official leaderboard
12 MiMo-V2.6 Pro 67.5% Sep 2026 Official leaderboard
13 GLM-5.3 66.7% Aug 2026 Official leaderboard
14 Muse Spark 1.2 66.5% Aug 2026 Official leaderboard
15 Grok 4.6 66% Aug 2026 Official leaderboard
16 DeepSeek V4.1 Flash 65.6% Sep 2026 Official leaderboard
17 Muse Spark 1.3 65% Sep 2026 Official leaderboard
18 Mistral Large 4 63.7% Oct 2026 Official leaderboard
19 Gemini 3.8 Flash 63.4% Sep 2026 Official leaderboard
20 GPT-6 Luna 62.5% Sep 2026 Official leaderboard
21 Claude Haiku 5.5 61.8% Oct 2026 Official leaderboard
22 DeepSeek V4 Pro 0813 61.7% Aug 2026 Official leaderboard
23 MiniMax-M3 57.5% Jun 2026 Official leaderboard
24 Gemini 3.5 Flash-Lite 55.7% Jul 2026 Official leaderboard

The best score, 75.4%, costs $2.11 a task

Runs on the cost frontier score higher than any run that's cheaper. Muse Wonder Full comes within 4.2 points of the best for $0.39 a task, 5 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Qwen Anthropic Xiaomi Corp OpenAI Meta Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Fable 5.1 Muse Wonder Full Qwen 3.8 Max MiMo-V2.6 Pro GPT-6 Luna 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Fable 5.1 Qwen 3.8 Max
ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5.1 75.4% $2.11
Claude Opus 5.5 72.1% $2.16
Muse Wonder Full 71.2% $0.39
Qwen 3.8 Max 70.1% $0.12
Muse Wonder 69.6% $0.30
Claude Sonnet 5.5 69% $1.16
Kimi K3 68.3% $0.18
Gemini 4 Argon 68.1% $0.51
Grok 4.7 68% $0.24
GPT-6.1 Sol 67.8% $0.23
GPT-6 Astra 67.7% $1.37
MiMo-V2.6 Pro 67.5% $0.02
GLM-5.3 66.7% $0.12
Muse Spark 1.2 66.5% $0.04
Grok 4.6 66% $0.10
DeepSeek V4.1 Flash 65.6% $0.05
Muse Spark 1.3 65% $0.05
Muse Spark 1.3 max 64.9% $0.06
Mistral Large 4 63.7% $0.15
Gemini 3.8 Flash 63.4% $0.17
GPT-6 Luna 62.5% $0.01
Claude Haiku 5.5 61.8% $0.09
DeepSeek V4 Pro 0813 61.7% $0.05
MiniMax-M3 57.5% $0.04
Gemini 3.5 Flash-Lite 55.7% $0.02

Number one over time

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

Anthropic
Own leaderboard 1 leader
Claude Fable 5.1 75.4%

Scores over time

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

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5.1 Qwen 3.8 Max MiniMax-M3 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5.1 Kimi K3

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

More in Law, finance and medicine

All 140 in Law, finance and medicine
  • 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%
  • 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 In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
  • KBL In use Models answer Korean legal knowledge and reasoning tasks and Korean bar exam questions. It has 7 knowledge tasks, 4 reasoning tasks and 2,510 bar exam items. Solar Pro 4 77.5%
  • 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. Solar Pro 4 93.2%
  • PRBench In use Model answers 1,100 expert-written law and finance questions. LLM judge grades against weighted expert rubrics of 10-30 criteria each. Muse Spark 1.3 61.6%