Vals Immigration Bench (Immigration Bench)
We show this table for reference; we do not rank on it.
Vals AI and Talent Mobility Fund benchmark of expert-written U.S. immigration law questions, scored against expert rubrics.
Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.
Rubric accuracy on Immigration Bench — October 28, 2026
We mirror the published rubric accuracy view for Immigration Bench. Claude Fable 5.1 leads the public snapshot at 75.44%, followed by Claude Opus 5.5 (72.14%) and Muse Wonder Full (71.22%). We do not use these results to rank models overall.
Claude Fable 5.1
Anthropic
Claude Opus 5.5
Anthropic
Muse Wonder Full
Meta
max reasoning
25 modelsKnowledgeCurrentDisplay onlyUpdated October 28, 2026
Rubric accuracy table (25 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Fable 5.1Anthropic | 75.44% | Not reported | Closed |
| 2 | Claude Opus 5.5Anthropic | 72.14% | Not reported | Closed |
| 3 | Muse Wonder FullMetamax reasoning | 71.22% | Not reported | Closed |
| 4 | Qwen3.8 MaxAlibaba | 70.14% | Not reported | Open |
| 5 | Muse WonderMetamax reasoning | 69.62% | Not reported | Closed |
| 6 | Claude Sonnet 5.5Anthropic | 68.96% | Not reported | Closed |
| 7 | Kimi K3Moonshot AImax reasoning | 68.26% | Not reported | Open |
| 8 | Gemini 4 ArgonGooglehigh reasoning | 68.08% | Not reported | Closed |
| 9 | Grok 4.7xAIxhigh reasoning | 68.05% | Not reported | Closed |
| 10 | GPT-6.1 SolOpenAImax reasoning | 67.81% | Not reported | Closed |
| 11 | GPT-6 AstraOpenAImax reasoning | 67.69% | Not reported | Closed |
| 12 | MiMo-V2.6-ProXiaomi | 67.50% | Not reported | Open |
| 13 | GLM-5.3Z.AImax reasoning | 66.73% | Not reported | Open |
| 14 | Muse Spark 1.2Metaxhigh reasoning | 66.54% | Not reported | Closed |
| 15 | Grok 4.6xAIhigh reasoning | 65.98% | Not reported | Closed |
| 16 | DeepSeek V4.1 Flash 1017DeepSeekmax reasoning | 65.64% | Not reported | Open |
| 17 | Muse Spark 1.3Metaxhigh reasoning | 65.00% | Not reported | Closed |
| 18 | Muse Spark 1.3 MaxMetamax reasoning | 64.93% | Not reported | Unknown |
| 19 | Mistral Large 4Mistralhigh reasoning | 63.71% | Not reported | Open |
| 20 | Gemini 3.8 FlashGooglehigh reasoning | 63.41% | Not reported | Closed |
| 21 | GPT-6 LunaOpenAImax reasoning | 62.53% | Not reported | Closed |
| 22 | Claude Haiku 5.5Anthropic | 61.79% | Not reported | Closed |
| 23 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 61.66% | Not reported | Open |
| 24 | MiniMax M3MiniMax | 57.53% | Not reported | Open |
| 25 | Gemini 3.5 Flash-LiteGooglehigh reasoning | 55.66% | Not reported | Closed |
How to read this leaderboard
Operator receipt: 25 sourced rows are currently displayable on this page; the leading published row is Claude Fable 5.1 at 75.44%.
Honest limit: Partial rubric credit is not a fully correct answer: Vals reports the top model fully correct on 26.3% of questions. Grading uses an AI judge, the test set is private, and the scope is professional and student pathways, not asylum or immigration court.
How Immigration Bench is shown here
BenchLM mirrors the public Vals AI Immigration Bench leaderboard captured from https://www.vals.ai/benchmarks/immigration and updated by Vals on October 28, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.
Immigration Bench 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
The published Immigration Bench snapshot places Claude Fable 5.1 first at 75.44%. The third row is 4.22 points behind. The broader top-10 range is 7.64 points, so many of the published results sit in a relatively narrow band.
25 models have been evaluated on Immigration Bench. 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. Immigration Bench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Immigration Bench
Year
2026
Tasks
262 private test questions
Format
Average share of expert rubric items met
Difficulty
Fact-specific U.S. immigration law guidance
Vals published Immigration Bench on October 28, 2026. Immigration experts wrote 519 questions from seven personas (international student, scholar or faculty, university advisor, employer, entrepreneur, self-petitioner and fragile or temporary status). Results cover the private 262-question test set; the other 257 questions are reserved for labs to license. Each question has an expert rubric, an AI grader marks each item met when the answer includes it, and the score is the share of rubric items met, averaged over questions. We mirror the published leaderboard as a display-only page, outside overall and category rankings.
Freshness and provenance
Version
Immigration Bench 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
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 Immigration Bench measure?
Vals AI and Talent Mobility Fund benchmark of expert-written U.S. immigration law questions, scored against expert rubrics.
Which model leads the published Immigration Bench snapshot?
Claude Fable 5.1 currently leads the published Immigration Bench snapshot with 75.44% rubric accuracy. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Immigration Bench?
The October 28, 2026 snapshot contains 25 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.