Benchmark explorer.
Humanity’s Last ExamAA Index 4.3.2 collection · Higher is better
Source ↗Method
Artificial Analysis evaluation. Index 4.3.2 collection.
Score × cost
Where performance meets price
Humanity’s Last Exam · 1M input + 250K output tokens · uncached USD
17 priced resultsPareto frontier
Anthropic: Claude Opus 5.5 ↗61.35% Humanity’s Last Exam · $9.00 workload · Amazon Bedrock · via OpenRouter
Evaluation: Effort: max · Max, Default Fallback. Price: $4.00 input / $20.00 output per 1M tokens · amazon-bedrock.
Numbers match the model key. Shared positions show an index range.
Model key Score / workload
- GLM-5.3-Flash39.85% / $0.124
- DeepSeek-V4.1-Flash-101743.33% / $0.135
- Qwen3.8-27B33.92% / $0.145
- OpenAI: GPT-6 Luna38.51% / $0.225
- GLM-5.342.26% / $0.429
- Reflection: Beam 1.0 501B A23B34.66% / $0.50
- Xiaomi: MiMo-V2.6-Pro49.35% / $0.5828
- Kimi-K346.9% / $1.55
- Step-5-Preview46.48% / $1.675
- Thinking Machines: Inkling31.88% / $1.9625
- Google: Gemini 4 Argon57.09% / $2.25
- Meta: Muse Spark Open43.88% / $2.3125
- Anthropic: Claude Sonnet 5.554.96% / $4.50
- OpenAI: GPT-6.1 Sol52.92% / $4.50
- Anthropic: Claude Opus 5.561.35% / $9.00
- OpenAI: GPT-6 Astra54.68% / $11.25
- Anthropic: Claude Fable 5.159.13% / $22.50
Pareto frontier
- GLM-5.3-FlashScore39.85%Workload cost$0.124Wafer · via OpenRouter · wafer/nvfp4 · quantization=nvfp4Evaluation: Effort: max. Price: $0.074 input / $0.20 output per 1M tokens · wafer/nvfp4 · quantization=nvfp4.
- DeepSeek-V4.1-Flash-1017Score43.33%Workload cost$0.135Decart · via OpenRouter · decart/fp4 · quantization=fp4Evaluation: Effort: max · Max. Price: $0.09 input / $0.18 output per 1M tokens · decart/fp4 · quantization=fp4.
- Xiaomi: MiMo-V2.6-ProScore49.35%Workload cost$0.5828DeepInfra · via OpenRouter · deepinfra/fp8 · quantization=fp8Evaluation: Effort not specified. Price: $0.387 input / $0.783 output per 1M tokens · deepinfra/fp8 · quantization=fp8.
- Google: Gemini 4 ArgonScore57.09%Workload cost$2.25Google AI Studio · via OpenRouter · google-ai-studio/flex · tier=flexEvaluation: Effort: high · High. Price: $1.00 input / $5.00 output per 1M tokens · google-ai-studio/flex · tier=flex.
- Anthropic: Claude Opus 5.5Score61.35%Workload cost$9.00Amazon Bedrock · via OpenRouter · amazon-bedrockEvaluation: Effort: max · Max, Default Fallback. Price: $4.00 input / $20.00 output per 1M tokens · amazon-bedrock.
Frontier: no result is cheaper with an equal or better score, or equally priced with a better score. Each point uses its lowest generally available workload price from one endpoint; evaluation and provider configurations remain visible.
| # | Model | Humanity’s Last Exam | Configuration | API rate / unit | Source |
|---|---|---|---|---|---|
| 1 | Anthropic: Claude Opus 5.5anthropic | 61.35% | Effort: max · Max, Default FallbackIndependent result | $4.00 in / $20.00 out/ 1M tokens · Amazon Bedrock · via OpenRouteramazon-bedrock | Result ↗Oct 21, 2026 |
| 2 | Anthropic: Claude Fable 5.1anthropic | 59.13% | Effort: max · Max, Default FallbackIndependent result | $10.00 in / $50.00 out/ 1M tokens · Amazon Bedrock · via OpenRouteramazon-bedrock | Result ↗Oct 21, 2026 |
| 3 | Google: Gemini 4 Argongoogle | 57.09% | Effort: high · HighIndependent result | $1.00 in / $5.00 out/ 1M tokens · Google AI Studio · via OpenRoutergoogle-ai-studio/flex · tier=flex | Result ↗Oct 21, 2026 |
| 4 | Anthropic: Claude Sonnet 5.5anthropic | 54.96% | Effort: max · Max, Default FallbackIndependent result | $2.00 in / $10.00 out/ 1M tokens · Amazon Bedrock · via OpenRouteramazon-bedrock | Result ↗Oct 21, 2026 |
| 5 | OpenAI: GPT-6 Astraopenai | 54.68% | Effort: max · MaxIndependent result | $5.00 in / $25.00 out/ 1M tokens · OpenAI · via OpenRouteropenai/flex · tier=flex · prompt ≤ 272000 tokens | Result ↗Oct 21, 2026 |
| 6 | Muse Wonder FullMeta | 53.61% | Effort: max · MaxIndependent result | See provider quotes | Result ↗Oct 21, 2026 |
| 7 | OpenAI: GPT-6.1 Solopenai | 52.92% | Effort: max · MaxIndependent result | $2.00 in / $10.00 out/ 1M tokens · ExperientialStandard list rate | Result ↗Oct 21, 2026 |
| 8 | Muse WonderMeta | 50.09% | Effort: max · MaxIndependent result | See provider quotes | Result ↗Oct 21, 2026 |
| 9 | Xiaomi: MiMo-V2.6-Proxiaomi | 49.35% | Effort not specifiedIndependent result | $0.387 in / $0.783 out/ 1M tokens · DeepInfra · via OpenRouterdeepinfra/fp8 · quantization=fp8 | Result ↗Oct 21, 2026 |
| 10 | MiniMax-M3.1-FlashMiniMax | 47.96% | Effort not specifiedIndependent result | See provider quotes | Result ↗Oct 21, 2026 |
| 11 | Gemini 3.8 FlashGoogle | 47.82% | Effort: high · HighIndependent result | See provider quotes | Result ↗Oct 21, 2026 |
| 12 | Kimi-K3moonshotai | 46.9% | Effort: max · MaxIndependent result | $0.238 in / $24.00 out/ 1M tokens · Wafer · via OpenRouterwafer · quantization=mxfp4 | Result ↗Oct 21, 2026 |
| 13 | Step-5-Previewstepfun-ai | 46.48% | Effort not specifiedIndependent result | $1.00 in / $2.70 out/ 1M tokens · DeepInfra | Result ↗Oct 21, 2026 |
| 14 | Claude Haiku 5.5Anthropic | 44.39% | Effort: max · MaxIndependent result | See provider quotes | Result ↗Oct 21, 2026 |
| 15 | Meta: Muse Spark Openmeta | 43.88% | Effort: xhigh · XhighIndependent result | $1.25 in / $4.25 out/ 1M tokens · baseten · via Hugging Face | Result ↗Oct 21, 2026 |
| 16 | DeepSeek-V4.1-Flash-1017deepseek-ai | 43.33% | Effort: max · MaxIndependent result | $0.07 in / $0.50 out/ 1M tokens · Relace · via OpenRouterrelace | Result ↗Oct 21, 2026 |
| 17 | Grok 4.7SpaceXAI | 43.14% | Effort: xhigh · XhighIndependent result | See provider quotes | Result ↗Oct 21, 2026 |
| 18 | Qwen3.8 Max (0902)Alibaba | 43.1% | Effort not specifiedIndependent result | See provider quotes | Result ↗Oct 21, 2026 |
| 19 | GLM-5.3zai-org | 42.26% | Effort: max · MaxIndependent result | $0.079 in / $1.40 out/ 1M tokens · Wafer · via OpenRouterwafer/us · quantization=fp4 | Result ↗Oct 21, 2026 |
| 20 | GLM-5.3-Flashzai-org | 39.85% | Effort: maxIndependent result | $0.06 in / $1.80 out/ 1M tokens · Reka · via OpenRouterreka | Result ↗Oct 21, 2026 |
| 21 | OpenAI: GPT-6 Lunaopenai | 38.51% | Effort: max · MaxIndependent result | $0.10 in / $0.50 out/ 1M tokens · Azure · via OpenRouterazure · prompt ≤ 272000 tokens | Result ↗Oct 21, 2026 |
| 22 | Mistral Large 4 PreviewMistral | 35.03% | Effort not specifiedIndependent result | See provider quotes | Result ↗Oct 21, 2026 |
| 23 | Reflection: Beam 1.0 501B A23Breflection | 34.66% | Effort not specifiedIndependent result | $0.25 in / $1.00 out/ 1M tokens · Nebius · via OpenRouternebius/fp8 · quantization=fp8 | Result ↗Oct 21, 2026 |
| 24 | Qwen3.8-27BQwen | 33.92% | Effort: xhigh · XhighIndependent result | $0.02 in / $0.50 out/ 1M tokens · Tenstorrent · via OpenRoutertenstorrent/fp4 · quantization=fp4 | Result ↗Oct 21, 2026 |
| 25 | K2 Horizon 375B A23BInstitute of Foundation Models | 31.97% | Effort not specifiedIndependent result | See provider quotes | Result ↗Oct 21, 2026 |
| 26 | Thinking Machines: Inklingthinkingmachines | 31.88% | Effort: xhigh · XhighIndependent result | $0.95 in / $4.05 out/ 1M tokens · deepinfra · via Hugging Face | Result ↗Oct 21, 2026 |
| 27 | Nemotron 3.5 Super VLNVIDIA | 30.72% | Effort not specifiedIndependent result | See provider quotes | Result ↗Oct 21, 2026 |