Scale Labs Humanity's Last Exam
We show this table for reference; we do not rank on it.
A Scale Labs public leaderboard mirrored as display-only reference data. It does not affect BenchLM rankings.
Scale score on Humanity's Last Exam — October 15, 2026 snapshot
We mirror the published scale score view for Humanity's Last Exam. GPT-6 Astra leads the public snapshot at 54.8%, followed by Fable 5.1 (xhigh)\n (46.5%) and gemini-3.1-pro-preview (thinking high) (46.4%). We do not use these results to rank models overall.
GPT-6 Astra
OpenAI
Fable 5.1 (xhigh)\n
Anthropic
gemini-3.1-pro-preview (thinking high)
55 modelsKnowledgeCurrentDisplay onlyUpdated October 15, 2026 snapshot
Scale score table (55 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 54.8% | Not reported | Closed |
| 2 | Fable 5.1 (xhigh)\nAnthropic | 46.5% | Not reported | Unknown |
| 3 | gemini-3.1-pro-preview (thinking high)Google | 46.4% | Not reported | Unknown |
| 4 | Sonnet 5.5 (xhigh)Anthropic | 46.1% | Not reported | Unknown |
| 5 | Gemini 3.8 FlashGoogle | 44.5% | Not reported | Closed |
| 6 | GPT-5.4 ProOpenAI | 44.3% | Not reported | Closed |
| 7 | Muse SparkMeta | 40.6% | Not reported | Closed |
| 8 | gemini-3-pro-previewGoogle | 37.5% | Not reported | Unknown |
| 9 | gpt-5.4-2026-03-05 (xhigh thinking)OpenAI | 36.2% | Not reported | Unknown |
| 10 | Claude Opus 4.7Anthropic | 36.2% | Not reported | Closed |
| 11 | claude-opus-4-6-thinking-maxAnthropic | 34.4% | Not reported | Unknown |
| 12 | gpt-5-pro-2025-10-06OpenAI | 31.6% | Not reported | Unknown |
| 13 | Claude Haiku 5.5Anthropic | 30.1% | Not reported | Closed |
| 14 | GPT-5.2OpenAI | 27.8% | Not reported | Closed |
| 15 | gpt-5-2025-08-07OpenAI | 25.3% | Not reported | Unknown |
| 16 | Claude Opus 4.5 ThinkingAnthropic | 25.2% | Not reported | Closed |
| 17 | Kimi K2.5Moonshot AI | 24.4% | Not reported | Open |
| 18 | gpt-5.1-thinkingOpenAI | 23.7% | Not reported | Unknown |
| 19 | gemini-2.5-pro-preview-06-05Google | 21.6% | Not reported | Unknown |
| 20 | o3 (high) (April 2025)OpenAI | 20.3% | Not reported | Unknown |
| 21 | GPT-5 miniOpenAI | 19.4% | Not reported | Closed |
| 22 | o3 (medium) (April 2025)OpenAI | 19.2% | Not reported | Unknown |
| 23 | claude-opus-4-6 (Non-Thinking)Anthropic | 19% | Not reported | Unknown |
| 24 | Gemini 2.5 Pro Experimental (March 2025)Google | 18.2% | Not reported | Unknown |
| 25 | o4-mini (high) (April 2025)OpenAI | 18.1% | Not reported | Unknown |
| 26 | Gemini 2.5 Pro Preview (May 06 2025)Google | 17.8% | Not reported | Unknown |
| 27 | o4-mini (medium) (April 2025)OpenAI | 14.3% | Not reported | Unknown |
| 28 | Claude Opus 4.5Anthropic | 14.2% | Not reported | Closed |
| 29 | Claude Sonnet 4.5 ThinkingAnthropic | 13.7% | Not reported | Closed |
| 30 | Gemini 2.5 Flash (April 2025)Google | 12.1% | Not reported | Unknown |
| 31 | claude-opus-4-1-20250805-thinkingAnthropic | 11.5% | Not reported | Unknown |
| 32 | Gemini 2.5 Flash Preview (May 2025)Google | 11.0% | Not reported | Unknown |
| 33 | Claude Opus 4 (Thinking)Anthropic | 10.7% | Not reported | Unknown |
| 34 | gemini-3.1-flash-lite-previewGoogle | 8.6% | Not reported | Unknown |
| 35 | glm-4p5Z.AI | 8.3% | Not reported | Unknown |
| 36 | glm-4p5-airZ.AI | 8.1% | Not reported | Unknown |
| 37 | o1-proOpenAI | 8.1% | Not reported | Closed |
| 38 | Claude 3.7 Sonnet (Thinking)Anthropic | 8.0% | Not reported | Unknown |
| 39 | o1 (December 2024)OpenAI | 8.0% | Not reported | Unknown |
| 40 | claude-opus-4-1-20250805Anthropic | 7.9% | Not reported | Unknown |
| 41 | Claude Sonnet 4 (Thinking)\nAnthropic | 7.8% | Not reported | Unknown |
| 42 | Claude Sonnet 4.5Anthropic | 7.5% | Not reported | Closed |
| 43 | gpt-5.1-instantOpenAI | 6.8% | Not reported | Unknown |
| 44 | Claude Opus 4Anthropic | 6.7% | Not reported | Unknown |
| 45 | Gemini 2.0 Flash Thinking (January 2025)Google | 6.6% | Not reported | Unknown |
| 46 | Llama 4 MaverickMeta | 5.7% | Not reported | Open |
| 47 | Claude Sonnet 4Anthropic | 5.5% | Not reported | Unknown |
| 48 | GPT 4.5 PreviewOpenAI | 5.4% | Not reported | Unknown |
| 49 | GPT-4.1OpenAI | 5.4% | Not reported | Closed |
| 50 | Gemini-1.5-Pro-002Google | 4.6% | Not reported | Unknown |
| 51 | Mistral Medium 3Mistral | 4.5% | Not reported | Closed |
| 52 | Nova ProAmazon | 4.4% | Not reported | Closed |
| 53 | Claude 3.5 Sonnet (October 2024)Anthropic | 4.1% | Not reported | Unknown |
| 54 | Nova LiteAmazon | 3.6% | Not reported | Unknown |
| 55 | GPT-4o (November 2024)OpenAI | 2.7% | Not reported | Unknown |
How to read this leaderboard
Compare the published configurations as complete evaluation systems. The source can combine a base model, agent scaffold, tools, budget, and inference setting in each result.
Operator receipt: 55 sourced rows are currently displayable on this page; the leading published row is GPT-6 Astra at 54.8%.
Honest limit: This Scale table is display-only context, not benchmark provenance or a weighted model-only comparison.
How BenchLM shows Humanity's Last Exam
BenchLM mirrors 55 published rows from Scale Labs’ public Humanity's Last Exam leaderboard, captured on October 15, 2026 snapshot.
The table is display only. It is useful context for a published agent or model configuration, but it does not enter BenchLM’s overall or category rankings.
The published Humanity's Last Exam snapshot places GPT-6 Astra first at 54.8%. The third row is 8.4 points behind. The broader top-10 range is 18.6 points, so the table still separates the published systems.
55 models have been evaluated on Humanity's Last Exam. 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. Humanity's Last Exam is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Humanity's Last Exam
Year
2026
Tasks
55 published rows
Format
Published Scale leaderboard score
Difficulty
External agent and model evaluation
BenchLM mirrors 55 published rows from the Humanity's Last Exam public table captured on October 15, 2026 snapshot.
Freshness and provenance
Version
Humanity's Last Exam 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 Humanity's Last Exam measure?
A Scale Labs public leaderboard mirrored as display-only reference data. It does not affect BenchLM rankings.
Which model leads the published Humanity's Last Exam snapshot?
GPT-6 Astra currently leads the published Humanity's Last Exam snapshot with 54.8% scale score. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Humanity's Last Exam?
The October 15, 2026 snapshot snapshot contains 55 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.