Humanity's Last Exam without tools (HLE w/o tools)
Tool-free variant of Humanity's Last Exam that isolates a model's raw frontier reasoning.
Top models on HLE w/o tools — October 29, 2026
As of October 29, 2026, Claude Opus 5.5 leads the HLE w/o tools leaderboard with 64.4% , followed by Claude Fable 5.1 (60.9%) and Claude Mythos 5 (59%).
Claude Opus 5.5
Anthropic
Claude Fable 5.1
Anthropic
Claude Mythos 5
Anthropic
51 modelsKnowledge7% of Knowledge reference weightCurrentUpdated October 29, 2026
Leaderboard (51 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Opus 5.5Anthropic | 64.4% | Not reported | Closed |
| 2 | Claude Fable 5.1Anthropic | 60.9% | Not reported | Closed |
| 3 | Claude Mythos 5Anthropic | 59% | Not reported | Closed |
| 4 | Claude Sonnet 5.5Anthropic | 56.9% | Not reported | Closed |
| 5 | Claude Opus 5Anthropic | 56.3% | Not reported | Closed |
| 6 | Muse Spark 1.1Meta | 52.2% | Not reported | Closed |
| 7 | Kimi K3.1 ProMoonshot AI | 51.1% | Not reported | Pending |
| 8 | Sakana Fugu-UltraSakana AI | 50% | Not reported | Closed |
| 9 | Claude Opus 4.8Anthropic | 49.8% | Not reported | Closed |
| 10 | Pareto 26.9Unbiased | 49% | Not reported | Closed |
| 11 | Sakana FuguSakana AI | 47.2% | Not reported | Closed |
| 12 | Claude Opus 4.7 (Adaptive)Anthropic | 46.9% | Not reported | Closed |
| 13 | Claude Haiku 5.5Anthropic | 45.9% | Not reported | Closed |
| 14 | Gemini 3.1 ProGoogle | 45.4% | Not reported | Closed |
| 15 | Ornith-1.5-397BOrnith AI | 44.6% | Not reported | Open |
| 16 | Qwen3.8 MaxAlibaba | 43.6% | Not reported | Open |
| 17 | Kimi K3Moonshot AI | 43.5% | Not reported | Open |
| 18 | MiniMax M3.1 FlashMiniMax | 43.4% | Not reported | Closed |
| 19 | Hy4 previewTencent | 43.4% | Not reported | Open |
| 20 | Claude Sonnet 5Anthropic | 43.2% | Not reported | Closed |
| 21 | GPT-5.5 ProOpenAI | 43.1% | Not reported | Closed |
| 22 | Muse SparkMeta | 42.8% | Not reported | Closed |
| 23 | GPT-5.4 ProOpenAI | 42.7% | Not reported | Closed |
| 24 | GPT-5.5OpenAI | 41.4% | Not reported | Closed |
| 25 | GLM-5.2Z.AI | 40.5% | Not reported | Open |
| 26 | Claude Opus 4.6Anthropic | 40% | Not reported | Closed |
| 27 | GPT-5.4OpenAI | 39.8% | Not reported | Closed |
| 28 | Beam 1.0 501B-A23B FP8Reflection AI | 36% | Not reported | Open |
| 29 | Beam 1.0 501B-A23BReflection AI | 36% | Not reported | Open |
| 30 | Qwen3.8-Flash-NextAlibaba | 35.9% | Not reported | Open |
| 31 | Beam 1.0 501B-A23B NVFP4Reflection AI | 34.8% | Not reported | Open |
| 32 | MiMo-V2.5-ProXiaomi | 34% | Not reported | Closed |
| 33 | Nemotron 3.5 Super VL 120B A12BNVIDIA | 31.8% | Not reported | Open |
| 34 | Inkling-SmallThinking Machines Lab | 31.6% | Not reported | Open |
| 35 | Grok 4.20xAI | 31.6% | Not reported | Closed |
| 36 | Nemotron 3.5 Super VL 120B A12B NVFP4NVIDIA | 31.4% | Not reported | Open |
| 37 | Qwen3.8-27BAlibaba | 30.8% | Not reported | Open |
| 38 | InklingThinking Machines Lab | 30% | Not reported | Open |
| 39 | Solar Open 2Upstage | 28.8% | Not reported | Open |
| 40 | Nemotron 3.5 Bolt 74B-A9B BF16NVIDIA | 28.7% | Not reported | Open |
| 41 | GPT-5.4 miniOpenAI | 28.2% | Not reported | Closed |
| 42 | Nemotron 3 UltraNVIDIA | 26.7% | Not reported | Open |
| 43 | Ornith-1.5-35B-A3BOrnith AI | 25.6% | Not reported | Open |
| 44 | Nemotron 3.5 Bolt 74B-A9BNVIDIA | 25.3% | Not reported | Open |
| 45 | GPT-5.4 nanoOpenAI | 24.3% | Not reported | Closed |
| 46 | Ornith-1.5-9BOrnith AI | 20.2% | Not reported | Open |
| 47 | Gemma 4 31BGoogle | 19.5% | Not reported | Open |
| 48 | Spark-X2.6-4BSparkLLM | 17.9% | Not reported | Open |
| 49 | Nemotron 3.5 Lightning 30B A3B NVFP4NVIDIA | 10.5% | Not reported | Open |
| 50 | Gemma 4 26B A4BGoogle | 8.7% | Not reported | Open |
| 51 | Gemma 4 12BGoogle | 5.2% | Not reported | Open |
According to BenchLM.ai, Claude Opus 5.5 leads the HLE w/o tools benchmark with a score of 64.4%, followed by Claude Fable 5.1 (60.9%) and Claude Mythos 5 (59%). The scores show moderate spread, with meaningful differences between the top tier and mid-tier models.
51 models have been evaluated on HLE w/o tools. The benchmark falls in the Knowledge category. BenchAlign v5.8 gives HLE w/o tools 7% of the Knowledge reference weight, so it moves the Knowledge leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.
About HLE w/o tools
Year
2026
Tasks
Expert-level questions
Format
Tool-free expert QA
Difficulty
Frontier expert level
This variant removes external tools so the score reflects pure model performance on frontier expert questions.
Freshness and provenance
Version
HLE w/o tools 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 HLE w/o tools measure?
Tool-free variant of Humanity's Last Exam that isolates a model's raw frontier reasoning.
Which model scores highest on HLE w/o tools?
Claude Opus 5.5 by Anthropic currently leads with a score of 64.4% on HLE w/o tools.
How many models are evaluated on HLE w/o tools?
51 AI models have published results on HLE w/o tools in the BenchLM catalog.
Compare top models on HLE w/o tools
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.