Vals RSI Index v1.1 (Vals RSI Index)
We show this table for reference; we do not rank on it.
Tests autonomous AI research on compression, language-model training, judge-harness engineering, and post-training under fixed compute and time budgets.
Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.
Reference-anchored RSI index on Vals RSI Index — October 15, 2026
We mirror the published reference-anchored rsi index view for Vals RSI Index. Claude Opus 5.5 leads the public snapshot at 37.31%, followed by Claude Fable 5.1 (36.09%) and Claude Opus 5 (33.02%). We do not use these results to rank models overall.
Claude Opus 5.5
Anthropic
max reasoning
Claude Fable 5.1
Anthropic
max reasoning
Claude Opus 5
Anthropic
max reasoning
25 modelsAgenticCurrentDisplay onlyUpdated October 15, 2026
Reference-anchored RSI index table (25 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Opus 5.5Anthropicmax reasoning | 37.31% | Not reported | Closed |
| 2 | Claude Fable 5.1Anthropicmax reasoning | 36.09% | Not reported | Closed |
| 3 | Claude Opus 5Anthropicmax reasoning | 33.02% | Not reported | Closed |
| 4 | Gemini 4 ArgonGooglehigh reasoning | 30.55% | Not reported | Closed |
| 5 | GPT-6 SolOpenAImax reasoning | 28.11% | Not reported | Closed |
| 6 | GPT-6 AstraOpenAImax reasoning | 27.07% | Not reported | Closed |
| 7 | GPT-6.1 SolOpenAImax reasoning | 26.62% | Not reported | Closed |
| 8 | Grok 4.6xAIxhigh reasoning | 25.46% | Not reported | Closed |
| 9 | Muse Wonder FullMetamax reasoning | 24.43% | Not reported | Closed |
| 10 | Claude Fable 5Anthropicmax reasoning | 24.32% | Not reported | Closed |
| 11 | Claude Opus 4.8Anthropicmax reasoning | 23.98% | Not reported | Closed |
| 12 | GPT-5.6 SolOpenAImax reasoning | 23.88% | Not reported | Closed |
| 13 | GLM-5.3Z.AImax reasoning | 22.91% | Not reported | Open |
| 14 | Qwen3.8 MaxAlibabamax reasoning | 21.76% | Not reported | Open |
| 15 | Kimi K3Moonshot AImax reasoning | 20.88% | Not reported | Pending |
| 16 | Grok 4.7xAIxhigh reasoning | 20.20% | Not reported | Closed |
| 17 | Gemini 3.8 FlashGooglehigh reasoning | 19.93% | Not reported | Closed |
| 18 | Muse Spark 1.3 MaxMetamax reasoning | 19.64% | Not reported | Unknown |
| 19 | Claude Opus 4.7Anthropicmax reasoning | 19.08% | Not reported | Closed |
| 20 | Gemini 3.7 FlashGooglehigh reasoning | 18.27% | Not reported | Closed |
| 21 | GPT-5.5OpenAIxhigh reasoning | 16.48% | Not reported | Closed |
| 22 | GPT-5.2OpenAIxhigh reasoning | 15.80% | Not reported | Closed |
| 23 | GPT-5.4OpenAIxhigh reasoning | 15.48% | Not reported | Closed |
| 24 | Claude Opus 4.5Anthropichigh reasoning | 14.82% | Not reported | Closed |
| 25 | Kimi K2.5 ThinkingMoonshot AImax reasoning | 6.85% | Not reported | Unknown |
How to read this leaderboard
Operator receipt: 25 sourced rows are currently displayable on this page; the leading published row is Claude Opus 5.5 at 37.31%.
Honest limit: The task references mix published results, human agreement, and a frontier model. Compute budgets and grading protocols differ across tasks, and the composite has no published confidence interval. It does not establish recursive self-improvement.
How Vals RSI Index is shown here
We mirror the Vals RSI Index v1.1 table, updated September 29, 2026, with its published overall index and four task scores. The tasks cover compression, language-model training, judge-harness engineering, and post-training. Each result combines a model with its provider-specific agent harness at the highest supported reasoning effort.
The index averages four reference-anchored log scores and expresses the result as a percentage. Zero marks each task's baseline, 50% marks a selected published, human, or frontier-model reference, and 100% marks the theoretical best. This is an index scale rather than a task success rate. The source does not publish confidence intervals for the composite.
Snapshot
The published Vals RSI Index snapshot places Claude Opus 5.5 first at 37.31%. The third row is 4.29 points behind. The broader top-10 range is 12.99 points, so the table still separates the published systems.
25 models have been evaluated on Vals RSI Index. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Vals RSI Index is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Vals RSI Index
Year
2026
Tasks
4 autonomous research tasks
Format
Mean reference-anchored log score, expressed as a percentage
Difficulty
AI research under fixed compute and time budgets
The published composite averages four reference-anchored log scores. On the percentage scale, 0% marks a task baseline, 50% matches a selected published, human, or frontier-model reference, and 100% marks the theoretical best. These are index points rather than task success rates. Models use provider-specific agent harnesses at their highest supported reasoning effort. We keep every published row display only and outside overall and category rankings.
Freshness and provenance
Version
Vals RSI Index 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 Vals RSI Index measure?
Tests autonomous AI research on compression, language-model training, judge-harness engineering, and post-training under fixed compute and time budgets.
Which model leads the published Vals RSI Index snapshot?
Claude Opus 5.5 currently leads the published Vals RSI Index snapshot with 37.31% reference-anchored rsi index. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Vals RSI Index?
The October 15, 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 5,600+ readers.
One email each week. Unsubscribe anytime.