CAIS AI Dashboard Text Capabilities Index (CAIS Text Leaderboard)
We show this table for reference; we do not rank on it.
A Center for AI Safety dashboard view summarizing text capabilities across HLE and TextQuests.
Text average on CAIS Text Leaderboard — Current two-component source snapshot
We mirror the published text average view for CAIS Text Leaderboard. Gemini 4 Argon leads the public snapshot at 66.5%, followed by Fable 5.2 (66.3%) and GPT-6 Astra (63.9%). We do not use these results to rank models overall.
Gemini 4 Argon
Fable 5.2
Anthropic
GPT-6 Astra
OpenAI
68 modelsCross-domainCurrentDisplay onlyUpdated Current two-component source snapshot
Text average table (68 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Gemini 4 ArgonGoogle | 66.5% | Not reported | Closed |
| 2 | Fable 5.2Anthropic | 66.3% | Not reported | Unknown |
| 3 | GPT-6 AstraOpenAI | 63.9% | Not reported | Closed |
| 4 | GPT-6.1 SolOpenAI | 61.4% | Not reported | Closed |
| 5 | Gemini 4 CarbonGoogle | 59.0% | Not reported | Closed |
| 6 | Opus 5.5Anthropic | 56.3% | Not reported | Unknown |
| 7 | Fable 5.1Anthropic | 55.6% | Not reported | Unknown |
| 8 | Fable 5Anthropic | 54.4% | Not reported | Unknown |
| 9 | Opus 5Anthropic | 52.2% | Not reported | Unknown |
| 10 | Gemini 3.8 FlashGoogle | 50.2% | Not reported | Closed |
| 11 | Gemini 3.7 FlashGoogle | 49.3% | Not reported | Closed |
| 12 | GPT-5.6 SolOpenAI | 48.8% | Not reported | Closed |
| 13 | Muse WonderMeta | 47.6% | Not reported | Closed |
| 14 | Kimi K3.1 ProMoonshot AI | 46.4% | Not reported | Open |
| 15 | GPT-5.5OpenAI | 46.3% | Not reported | Closed |
| 16 | Gemini 3.1 ProGoogle | 45.8% | Not reported | Closed |
| 17 | Grok 4.6xAI | 43.5% | Not reported | Closed |
| 18 | GPT-5.4OpenAI | 41.3% | Not reported | Closed |
| 19 | Opus 4.8Anthropic | 41.2% | Not reported | Unknown |
| 20 | Kimi K3Moonshot AI | 40.5% | Not reported | Open |
| 21 | Gemini 3 ProGoogle | 39.7% | Not reported | Closed |
| 22 | Gemini 3.5 FlashGoogle | 39.1% | Not reported | Closed |
| 23 | GPT-5.6 TerraOpenAI | 38.2% | Not reported | Closed |
| 24 | Muse Spark 1.3Meta | 38.1% | Not reported | Closed |
| 25 | Opus 4.7Anthropic | 38.0% | Not reported | Unknown |
| 26 | Opus 4.6Anthropic | 37.5% | Not reported | Unknown |
| 27 | GLM-5.2Z.AI | 37.4% | Not reported | Open |
| 28 | Muse Spark 1.1Meta | 37.2% | Not reported | Closed |
| 29 | Gemini 3 FlashGoogle | 36.5% | Not reported | Closed |
| 30 | Muse Spark OpenMeta | 36.3% | Not reported | Open |
| 31 | GLM-5.3Z.AI | 35.6% | Not reported | Open |
| 32 | GPT-5.2OpenAI | 33.5% | Not reported | Closed |
| 33 | Opus 4.5Anthropic | 32.3% | Not reported | Unknown |
| 34 | Grok 4.5xAI | 32.3% | Not reported | Closed |
| 35 | Kimi K2.6Moonshot AI | 31.4% | Not reported | Open |
| 36 | GPT-5.6 LunaOpenAI | 31.3% | Not reported | Closed |
| 37 | GPT-5.1OpenAI | 30.7% | Not reported | Closed |
| 38 | GPT-5 (high)OpenAI | 27.7% | Not reported | Closed |
| 39 | GLM-5.1Z.AI | 27.6% | Not reported | Open |
| 40 | GPT-5.4-miniOpenAI | 26.6% | Not reported | Unknown |
| 41 | DeepSeek V4 Pro 0813DeepSeek | 26.4% | Not reported | Open |
| 42 | Sonnet 4.6Anthropic | 26.3% | Not reported | Unknown |
| 43 | Grok 4xAI | 26.2% | Not reported | Closed |
| 44 | o3OpenAI | 25.6% | Not reported | Closed |
| 45 | Kimi K2.5Moonshot AI | 24.9% | Not reported | Open |
| 46 | Grok 4.2xAI | 24.3% | Not reported | Unknown |
| 47 | Grok 4.3xAI | 23.4% | Not reported | Closed |
| 48 | Sonnet 4.5Anthropic | 22.9% | Not reported | Unknown |
| 49 | Gemini 2.5 ProGoogle | 22.4% | Not reported | Closed |
| 50 | DeepSeek V3.2 (Thinking)DeepSeek | 21.5% | Not reported | Open |
| 51 | Kimi K2Moonshot AI | 19.9% | Not reported | Closed |
| 52 | Grok 4 Fast (Reasoning)xAI | 19.0% | Not reported | Closed |
| 53 | Gemini 2.5 Pro-ExpGoogle | 18.8% | Not reported | Unknown |
| 54 | GPT-5-miniOpenAI | 18.5% | Not reported | Unknown |
| 55 | Gemini 3.1 Flash-LiteGoogle | 18.1% | Not reported | Closed |
| 56 | Grok 4.1 Fast (Reasoning)xAI | 17.4% | Not reported | Closed |
| 57 | Sonnet 4Anthropic | 17.1% | Not reported | Unknown |
| 58 | GPT-5.4-NanoOpenAI | 16.1% | Not reported | Unknown |
| 59 | o4-mini (high)OpenAI | 15.4% | Not reported | Closed |
| 60 | Gemini 2.5 FlashGoogle | 13.2% | Not reported | Closed |
| 61 | o3-miniOpenAI | 12.7% | Not reported | Closed |
| 62 | Haiku 4.5Anthropic | 12.4% | Not reported | Unknown |
| 63 | DeepSeek-R1DeepSeek | 11.8% | Not reported | Open |
| 64 | Gemini 2.5 Flash-LiteGoogle | 9.2% | Not reported | Closed |
| 65 | Sonnet 3.7Anthropic | 8.0% | Not reported | Unknown |
| 66 | o1OpenAI | 8.0% | Not reported | Closed |
| 67 | GPT-4oOpenAI | 7.9% | Not reported | Closed |
| 68 | GPT-5-NanoOpenAI | 6.6% | Not reported | Unknown |
How the CAIS Text Leaderboard is shown here
We mirror the current CAIS AI Dashboard text view as an average of each model's available HLE and TextQuests scores. Missing values are excluded from the average.
The current source registered two text components on October 5, 2026. The earlier four-component view also included ARC-AGI-2 and SWE-Bench Pro, so those averages are not comparable. We keep this composite outside weighted rankings.
Snapshot
The published CAIS Text Leaderboard snapshot places Gemini 4 Argon first at 66.5%. The third row is 2.6 points behind. The broader top-10 range is 16.3 points, so the table still separates the published systems.
68 models have been evaluated on CAIS Text Leaderboard. The benchmark falls in the Cross-domain category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. CAIS Text Leaderboard is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About CAIS Text Leaderboard
Year
2025
Tasks
HLE and TextQuests
Format
Average component score
Difficulty
Composite frontier text capability
On October 5, 2026, the current CAIS AI Dashboard registered HLE and TextQuests as its text components. The displayed average uses each model's available component scores and excludes missing values. The earlier four-component view also included ARC-AGI-2 and SWE-Bench Pro; these averages are not comparable. This composite remains outside weighted rankings.
Freshness and provenance
Version
CAIS Text Leaderboard 2025
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 CAIS Text Leaderboard measure?
A Center for AI Safety dashboard view summarizing text capabilities across HLE and TextQuests.
Which model leads the published CAIS Text Leaderboard snapshot?
Gemini 4 Argon currently leads the published CAIS Text Leaderboard snapshot with 66.5% text average. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on CAIS Text Leaderboard?
The Current two-component source snapshot snapshot contains 68 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 7,000+ subscribers.
One email each week. Unsubscribe anytime.