HLE With Tools: Humanity's Last Exam — With Tools
Humanity's Last Exam expert-level benchmark measured with tool access. Kept separate from no-tools and GPT-5.5 Pro-compute variants.
Models ranked
16
tracked on this benchmark
Score band
70.0 – 11.9
best → lowest tracked
Snapshot trend
+43.85
Oct 19 → Oct 26 · 2 models
Leaderboard
Tracked models ranked by Accuracy (higher is better).
Notes: Moonshot self-reported launch table, row 'HLE-Full w/ tools', column 'Kimi K3.1 Pro (max)'. Settings: reasoning effort max, temperature 1.0, top-p 0.95 (blog footnote). Full 2,500-question set.
Configuration: Muse Spark 1.1 (xhigh)
Notes: Muse Spark 1.1 model page benchmark table (raw HTML parsed). Methodology: evaluation report https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report/ (2026-07-09). All Muse Spark 1.1 results run via Meta Model API at xhigh reasoning effort (report §5.1, p.100-101). Report §5.2.5: full 2,500-question set, official HLE protocol with o3-mini judge. Retrieved 2026-10-17.
Notes: Meta HF model card Benchmarks table, xhigh reasoning effort.
Configuration: DeepSeek-V4-Pro-0813 (max effort)
Notes: DeepSeek first-party HF card DeepSeek-V4-Pro-0813, benchmark table (column 1 = DeepSeek-V4-Pro-0813). Note 1: code-agent tasks run with DeepSeek Harness minimal mode, max effort, temperature 1.0, top_p 0.95. Row 'HLE (wo / w tools)': 42.7 / 60.0. Tool set not detailed on the card. Retrieved 2026-10-26.
Notes: Official Anthropic launch benchmark.
Notes: Model variant: Kimi K3 (max reasoning effort). Benchmark variant: HLE-Full with tools. Harness/evaluator not disclosed by Moonshot. Source posture: Moonshot/Kimi self-reported launch table; confidence: medium because the value is vendor-reported and not independently reproduced. Recommended seed value: 56.
Configuration: Deep Research Max (preview 04-2026)
Notes: Transcribed from eval chart https://storage.googleapis.com/gweb-uniblog-publish-prod/documents/gemini-3.1-pro_deep-research-and-max_blog_evals.png (downloaded, img/drmax-evals.png). Blog: all results use web search, evaluated by Google DeepMind via public APIs. Full 2,500-question set incl. multimodal. Retrieved 2026-10-17.
Notes: Standard with-tools variant kept separate from GPT-5.5 Pro-compute HLE.
Configuration: DeepSeek-V4-Flash-0731 (max effort)
Notes: DeepSeek first-party HF card DeepSeek-V4-Pro-0813, column 2 = DeepSeek-V4-Flash-0731, row 'HLE (wo / w tools)': 37.8 / 51.5. The Flash-0731 card itself has no HLE row; tool set not detailed. Retrieved 2026-10-26.
Notes: Official Intern Science model card reports Agents-A1 at 47.6 on Humanity's Last Exam with tools. Evaluator/source: Intern Science self-reported model-card/project benchmark table. Public harness details are not fully reproducible from the seed row. Confidence: medium for reported value, low-medium for reproducibility. Recommended seed value: 47.6.
Configuration: DiffusionGemma 26B A4B IT (Entropy-Bounded sampler)
Notes: Google HF model card evaluation table (instruction-tuned model, recommended Entropy-Bounded sampler; column 'DiffusionGemma 26B A4B'). Row 'HLE with search'. Retrieved 2026-10-19.
How to read this benchmark
This benchmark scores models where higher is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.
Trust this score when
- There is a fresh timestamped snapshot (or multiple snapshots) for this benchmark.
- The model list covers the same version family you can actually deploy today.
- Top candidates overlap with your required routing and feature requirements.
Be cautious when
- There is only one benchmark snapshot or the dataset appears stale.
- The benchmark metric direction is opposite of your decision objective.
- The score difference between options is narrow and likely within implementation variance.
Related benchmarks
Last reviewed: Jun 17, 2026