Best AI model for reasoning
The best AI model for reasoning on current benchmarks is Claude Opus 5.5 (Anthropic): as of October 19, 2026 it has the highest Humanity's Last Exam (tools) score TensorFeed tracks, 67.7% (vendor-reported), 2.7 points ahead of Claude Fable 5.1 (65%, vendor-reported), among 16 models scored. On GPQA Diamond, GPT-6 Astra leads (96%, vendor-reported).
Ranked on Humanity's Last Exam (tools). Benchmark data as of October 19, 2026; prices as of October 19, 2026.
Humanity's Last Exam (tools): 67.7% (vendor-reported). $4.00 / $20.00 per 1M tokens.
Humanity's Last Exam (tools): 63.9% (vendor-reported). $0.30 / $1.20 per 1M tokens.
Ranking on Humanity's Last Exam (tools)
Each score links to its source; vendor-reported and independent figures are labeled and are not interchangeable.
| # | Model | Provider | Humanity's Last Exam (tools) | Price in / out | TFII | GPQA Diamond |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | Anthropic | 67.7%vendor | $4.00 / $20.00 | 97.1 | - |
| 2 | Claude Fable 5.1 | Anthropic | 65%vendor | $10.00 / $50.00 | 89.9 | 93.43% |
| 3 | Claude Sonnet 5.5 | Anthropic | 64.5%vendor | $2.00 / $10.00 | 93.7 | - |
| 4 | DeepSeek V4.1 Flash | DeepSeek | 63.9%vendor | $0.30 / $1.20 | - | 90.9% |
| 5 | Claude Fable 5 | Anthropic | 63.9% | $10.00 / $50.00 | 81.6 | 93.18% |
| 6 | Claude Opus 5 | Anthropic | 63.6%vendor | $5.00 / $25.00 | 83.9 | 93.43% |
| 7 | GLM-5.3 | Z.ai | 62.5%vendor | $1.40 / $4.40 | 71.8 | 88.13% |
| 8 | Claude Opus 4.8 | Anthropic | 57.9% | $5.00 / $25.00 | 67.9 | 93.6% |
| 9 | Claude Sonnet 5 | Anthropic | 57.4%vendor | $2.00 / $10.00 | 59.2 | 88.89% |
| 10 | GPT-6 Astra | OpenAI | 57.2%vendor | $10.00 / $50.00 | 83.3 | 96% |
| 11 | Qwen3.8 2.4T-A95B | Alibaba | 56.2%vendor | $2.00 / $6.00 | - | 92.6% |
| 12 | Qwen3.8-Max | Alibaba | 56.2%vendor | $2.00 / $6.00 | 77.4 2024 set | 92.6% |
| 13 | Kimi K3 | Moonshot AI | 56%vendor | $3.00 / $15.00 | 65.1 | 93.5% |
| 14 | GLM-5.3 Flash | Z.ai | 55.3%vendor | $0.15 / $0.50 | 62.8 | 86.36% |
| 15 | GLM-5.2 | Z.ai | 54.7%vendor | $1.40 / $4.40 | 75.6 2024 set | 91.2% |
| 16 | Inkling | Thinking Machines | 46%vendor | - | 35.7 | 87.2% |
Prices are list prices per 1M input / output tokens. TFII is the headline TensorFeed Intelligence Index score; scores marked 2024 set come from older benchmarks and are not comparable with the rest.
How this is ranked
Humanity's Last Exam was built to resist the saturation that overtook MMLU: multidisciplinary, expert-level questions where the 2026 frontier still spreads out. The with-tools mode is how agents run in production. GPQA Diamond, graduate-level science, sits alongside; its leaders now cluster within about a point of each other.
Every Humanity's Last Exam (tools) figure is vendor-reported, and because the question set is public, search access can reach discussion of the questions themselves. Read this column as an upper bound for a model plus its retrieval stack, not a measure of what the weights know.
The full board, scoring notes and score ranges are on the Humanity's Last Exam (tools) leaderboard. The same scores are free as JSON at /api/benchmarks.
Questions
What is the best AI model for reasoning?
The best AI model for reasoning on current benchmarks is Claude Opus 5.5 (Anthropic): as of October 19, 2026 it has the highest Humanity's Last Exam (tools) score TensorFeed tracks, 67.7% (vendor-reported), 2.7 points ahead of Claude Fable 5.1 (65%, vendor-reported), among 16 models scored. On GPQA Diamond, GPT-6 Astra leads (96%, vendor-reported).
Which of the top AI models for reasoning is cheapest?
Of the five highest Humanity's Last Exam (tools) scores TensorFeed tracks, the lowest-priced model with a list price is DeepSeek V4.1 Flash (63.9%, vendor-reported), at $0.30 input and $1.20 output per 1M tokens as of October 19, 2026.
Why rank AI models for reasoning on Humanity's Last Exam (tools)?
Humanity's Last Exam was built to resist the saturation that overtook MMLU: multidisciplinary, expert-level questions where the 2026 frontier still spreads out. The with-tools mode is how agents run in production. GPQA Diamond, graduate-level science, sits alongside; its leaders now cluster within about a point of each other.