Humanity's Last Exam with tools (HLE w/ tools)
Tool-augmented Humanity's Last Exam scores reported in DeepSeek-V4 thinking-mode evaluations.
Top models on HLE w/ tools — November 18, 2026
As of November 18, 2026, Kimi K3.1 Pro leads the HLE w/ tools leaderboard with 70.0% , followed by Claude Opus 5.5 (67.7%) and Claude Opus 5 (64.7%).
Kimi K3.1 Pro
Moonshot AI
Reported by Moonshot AIClaude Opus 5.5
Anthropic
Reported by AnthropicClaude Opus 5
Anthropic
Reported by Anthropic39 modelsAgentic3% of Agentic reference weightCurrentUpdated November 18, 2026
Leaderboard (39 models)
All 39 results on this page were reported by the labs that built the models.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Kimi K3.1 ProMoonshot AIReported by Moonshot AI | 70.0% | Reported by Moonshot AI | Not reported | Open |
| 2 | Claude Opus 5.5AnthropicReported by Anthropic | 67.7% | Reported by Anthropic | Not reported | Closed |
| 3 | Claude Opus 5AnthropicReported by Anthropic | 64.7% | Reported by Anthropic | Not reported | Closed |
| 4 | Claude Sonnet 5.5AnthropicReported by Anthropic | 64.5% | Reported by Anthropic | Not reported | Closed |
| 5 | DeepSeek V4.1 Flash 1017DeepSeekReported by DeepSeek | 63.9% | Reported by DeepSeek | Not reported | Open |
| 6 | LongCat-2.5MeituanReported by Meituan | 63.1% | Reported by Meituan | Not reported | Open |
| 7 | GLM-5.3Z.AIReported by Z.AI | 62.5% | Reported by Z.AI | Not reported | Open |
| 8 | Muse Spark OpenMetaReported by Meta | 61.0% | Reported by Meta | Not reported | Open |
| 9 | DeepSeek V4 ProDeepSeekReported by DeepSeek | 60.0% | Reported by DeepSeek | Not reported | Open |
| 10 | Hy4TencentReported by Tencent | 59.1% | Reported by Tencent | Not reported | Pending |
| 11 | Claude Sonnet 5AnthropicReported by Anthropic | 57.4% | Reported by Anthropic | Not reported | Closed |
| 12 | Claude Haiku 5.5AnthropicReported by Anthropic | 57.4% | Reported by Anthropic | Not reported | Closed |
| 13 | GPT-6 AstraOpenAIReported by OpenAI | 57.2% | Reported by OpenAI | Not reported | Closed |
| 14 | Ling 3.0 Flash SanteInclusionAIReported by InclusionAI | 57.1% | Reported by InclusionAI | Not reported | Open |
| 15 | Qwen3.8 MaxAlibabaReported by Alibaba | 56.2% | Reported by Alibaba | Not reported | Open |
| 16 | Apodex 1.1ApodexReported by Apodex | 56.1% | Reported by Apodex | Not reported | Closed |
| 17 | Ornith-1.5-397BOrnith AIReported by Ornith AI | 56.1% | Reported by Ornith AI | Not reported | Open |
| 18 | Hy4 previewTencentReported by Tencent | 55.4% | Reported by Tencent | Not reported | Open |
| 19 | GLM-5.3-FlashZ.AIReported by Z.AI | 55.3% | Reported by Z.AI | Not reported | Open |
| 20 | Hy4 flashTencentReported by Tencent | 55.2% | Reported by Tencent | Not reported | Pending |
| 21 | Sage-A1.5InternScienceReported by InternScience | 55.1% | Reported by InternScience | Not reported | Open |
| 22 | GLM-5.2Z.AIReported by Z.AI | 54.7% | Reported by Z.AI | Not reported | Open |
| 23 | Qwen3.7 MaxAlibabaReported by Alibaba | 53.5% | Reported by Alibaba | Not reported | Closed |
| 24 | dots3-note PreviewDots StudioReported by Dots Studio | 52.6% | Reported by Dots Studio | Not reported | Open |
| 25 | GLM-5.1Z.AIReported by Z.AI | 52.3% | Reported by Z.AI | Not reported | Open |
| 26 | Beam 1.0 501B-A23B FP8Reflection AIReported by Reflection AI | 51.1% | Reported by Reflection AI | Not reported | Open |
| 27 | Beam 1.0 501B-A23B NVFP4Reflection AIReported by Reflection AI | 51.1% | Reported by Reflection AI | Not reported | Open |
| 28 | Beam 1.0 501B-A23BReflection AIReported by Reflection AI | 51.1% | Reported by Reflection AI | Not reported | Open |
| 29 | GLM-5Z.AIReported by Z.AI | 50.4% | Reported by Z.AI | Not reported | Open |
| 30 | Agents-A1InternScienceReported by InternScience | 47.6% | Reported by InternScience | Not reported | Open |
| 31 | Step 3.7 FlashStepFunReported by StepFun | 47.2% | Reported by StepFun | Not reported | Open |
| 32 | DeepSeek V4 Flash 0731DeepSeekReported by DeepSeek | 45.1% | Reported by DeepSeek | Not reported | Open |
| 33 | DeepSeek V4 Pro (High)DeepSeekReported by DeepSeek | 44.7% | Reported by DeepSeek | Not reported | Open |
| 34 | DeepSeek V4 Flash (High)DeepSeekReported by DeepSeek | 40.3% | Reported by DeepSeek | Not reported | Open |
| 35 | Nemotron 3 UltraNVIDIAReported by NVIDIA | 37.4% | Reported by NVIDIA | Not reported | Open |
| 36 | Spark-X2.6-4BSparkLLMReported by SparkLLM | 36.9% | Reported by SparkLLM | Not reported | Open |
| 37 | Ornith-1.5-35B-A3BOrnith AIReported by Ornith AI | 33.4% | Reported by Ornith AI | Not reported | Open |
| 38 | Ornith-1.5-9BOrnith AIReported by Ornith AI | 30.5% | Reported by Ornith AI | Not reported | Open |
| 39 | LFM3-1.7BLiquidAIReported by LiquidAI | 13.9% | Reported by LiquidAI | Not reported | Open |
- Lab-reported
- Benchmark-run
- Independent
Results reported by model developersUnranked
Labs ran these themselves, with a metric, split or setting the table above does not use. They take no rank here.
| Model | Reported value | What was measured | Against the table above | Who ran it |
|---|---|---|---|---|
| Apodex 1.1Apodex | 53.2Score | ReAct single agentThe stored row uses Agent Team; with tools | Not comparable to the scores above | Reported by Apodex |
| Step 3.7 FlashStepFun | 48.1Score | Model card prose figureThe tool setup is not stated, and it differs from the stored launch-page value | Not comparable to the scores above | Reported by StepFun |
A lab’s own result is shown for reference. It never takes a rank on this board.
How we label sourcesAccording to BenchLM.ai, Kimi K3.1 Pro leads the HLE w/ tools benchmark with a score of 70.0%, followed by Claude Opus 5.5 (67.7%) and Claude Opus 5 (64.7%). The scores show moderate spread, with meaningful differences between the top tier and mid-tier models.
39 models have been evaluated on HLE w/ tools. The benchmark falls in the Agentic category. BenchAlign v5.8 gives HLE w/ tools 3% of the Agentic reference weight, so it moves the Agentic leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.
About HLE w/ tools
Year
2026
Tasks
Expert questions with tool use
Format
Pass@1
Difficulty
Frontier tool-augmented reasoning
BenchLM stores HLE with tools as a provider-table row when a provider publishes exact values.
Freshness and provenance
Version
HLE w/ 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/ tools measure?
Tool-augmented Humanity's Last Exam scores reported in DeepSeek-V4 thinking-mode evaluations.
Which model scores highest on HLE w/ tools?
Kimi K3.1 Pro by Moonshot AI currently leads with a score of 70.0% on HLE w/ tools.
How many models are evaluated on HLE w/ tools?
39 AI models have published results on HLE w/ tools in the BenchLM catalog.
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.