Coding
Terminal-Bench 4.0 in the Artificial Analysis evaluation harness.
See model strengths as a shape, not a slogan.
Five task lenses from the same independent AA v4.3.2 snapshot. Each axis places the current benchmark leader at 100; missing scores remain empty.
Evidence state is part of the score.
This chart uses gpt-college-aa-v4.3.2-relative-v2. Values are relative benchmark scores, not percentages or a comprehensive capability assessment. Each capability index = raw score / current benchmark leader × 100. Overall = the average of all five indices, rounded to one decimal; requires all five scores.
Gemini 4 ArgonGoogle · 5 measured · 0 provisional axes
Claude Opus 5.5Anthropic · 5 measured · 0 provisional axes
Claude Sonnet 5.5Anthropic · 5 measured · 0 provisional axes
GPT-6.1 SolOpenAI · 5 measured · 0 provisional axes
GPT-6 SolOpenAI · 5 measured · 0 provisional axes
GPT-6 LunaOpenAI · 5 measured · 0 provisional axes
Grok 4.7xAI · 5 measured · 0 provisional axes
Grok 4.6xAI · 5 measured · 0 provisional axes
Qwen3.8-MaxAlibaba Cloud · 5 measured · 0 provisional axes
Qwen3.8-Max-0902Alibaba Cloud · 5 measured · 0 provisional axes
Qwen3.8-Flash-NextAlibaba Cloud · 0 measured · 0 provisional axes
GLM-5.3Z.ai · 5 measured · 0 provisional axes
MiniMax M3MiniMax · 5 measured · 0 provisional axes
Mistral Medium 3.5Mistral · 5 measured · 0 provisional axes
Claude Sonnet 5Anthropic · 0 measured · 0 provisional axes
Claude Haiku 4.5Anthropic · 0 measured · 0 provisional axes
Claude Mythos 5.1Anthropic · 0 measured · 0 provisional axes
Gemini 3.8 Flash CyberGoogle · 0 measured · 0 provisional axes
GPT-5.6 SolOpenAI · 0 measured · 0 provisional axes
GPT-5.6 TerraOpenAI · 0 measured · 0 provisional axes
GPT-5.6 LunaOpenAI · 0 measured · 0 provisional axes
Gemini 3.5 FlashGoogle · 0 measured · 0 provisional axes
Gemini 3.6 FlashGoogle · 0 measured · 0 provisional axes
Claude Fable 5Anthropic · 0 measured · 0 provisional axes
Claude Opus 5Anthropic · 0 measured · 0 provisional axes
Claude Opus 4.8Anthropic · 0 measured · 0 provisional axes
Claude Opus 4.7Anthropic · 0 measured · 0 provisional axes
Kimi K3Moonshot AI · 5 measured · 0 provisional axes
DeepSeek V4 ProDeepSeek · 0 measured · 0 provisional axes
DeepSeek V4 FlashDeepSeek · 0 measured · 0 provisional axes
OpenAI GPT-5.4OpenAI · 0 measured · 0 provisional axes
OpenAI GPT-5.2 ThinkingOpenAI · 0 measured · 0 provisional axes
Claude Opus 4.6Anthropic · 0 measured · 0 provisional axes
Gemini 3.1 ProGoogle · 0 measured · 0 provisional axes
Gemini 3 ProGoogle · 0 measured · 0 provisional axes
DeepSeek-R1DeepSeek · 0 measured · 0 provisional axes
Gemini 3.7 FlashGoogle · 0 measured · 0 provisional axes
GLM-5.3-FlashZ.ai · 5 measured · 0 provisional axes
Claude Fable 5.1Anthropic · 5 measured · 0 provisional axes
Gemini 3.8 FlashGoogle · 5 measured · 0 provisional axes
Qwen3.8-FlashAlibaba Cloud · 0 measured · 0 provisional axes
GPT-6 AstraOpenAI · 5 measured · 0 provisional axes
DeepSeek-V3DeepSeek · 0 measured · 0 provisional axes
Tencent Hunyuan-LargeTencent · 0 measured · 0 provisional axes
Hunyuan-T1Tencent · 0 measured · 0 provisional axes
DeepSeek-V4.1-FlashDeepSeek · 5 measured · 0 provisional axes
Task Lenses
Each lens uses one evaluation and one harness. Open its ranking to see original percentages or Elo scores.
Reasoning
Humanity's Last Exam as evaluated by Artificial Analysis; the source configuration is preserved.
Research
GDPval-AA v2.1 professional knowledge-work Elo. This is a task lens, not an assessment of every research skill.
Multimodal
GDP.pdf professional document understanding, including visual tables, charts and diagrams; not an audio/video score.
Agentic
AutomationBench-AA multi-step business workflows in the Artificial Analysis harness.
Capability Matrix
| Model | Coding | Reasoning | Research | Multimodal | Agentic | Coverage |
|---|---|---|---|---|---|---|
| Gemini 4 ArgonGoogle | 89.1Measured | 93.4Measured | 87.3Measured | 71.0Measured | 100.0Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Claude Opus 5.5Anthropic | 93.8Measured | 100.0Measured | 100.0Measured | 83.9Measured | 89.7Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Claude Sonnet 5.5Anthropic | 100.0Measured | 90.2Measured | 99.9Measured | 83.9Measured | 91.0Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| GPT-6.1 SolOpenAI | 87.5Measured | 86.9Measured | 85.3Measured | 100.0Measured | 83.3Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| GPT-6 SolOpenAI | 68.8Measured | 78.7Measured | 81.5Measured | 80.6Measured | 79.5Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| GPT-6 LunaOpenAI | 20.3Measured | 63.9Measured | 77.9Measured | 74.2Measured | 67.9Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Grok 4.7xAI | 40.6Measured | 70.5Measured | 91.8Measured | 64.5Measured | 84.6Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Grok 4.6xAI | 32.8Measured | 70.5Measured | 87.2Measured | 54.8Measured | 85.9Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Qwen3.8-MaxAlibaba Cloud | 29.7Measured | 70.5Measured | 87.1Measured | 64.5Measured | 62.8Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Qwen3.8-Max-0902Alibaba Cloud | 60.9Measured | 70.5Measured | 90.6Measured | 74.2Measured | 71.8Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Qwen3.8-Flash-NextAlibaba Cloud | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| GLM-5.3Z.ai | 65.6Measured | 68.9Measured | 89.7Measured | 35.5Measured | 79.5Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| MiniMax M3MiniMax | 3.1Measured | 63.9Measured | 67.5Measured | 32.3Measured | 26.9Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Mistral Medium 3.5Mistral | 0.0Measured | 23.0Measured | 40.5Measured | 9.7Measured | 7.7Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Claude Sonnet 5Anthropic | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Claude Haiku 4.5Anthropic | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Claude Mythos 5.1Anthropic | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Gemini 3.8 Flash CyberGoogle | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| GPT-5.6 SolOpenAI | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| GPT-5.6 TerraOpenAI | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| GPT-5.6 LunaOpenAI | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Gemini 3.5 FlashGoogle | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Gemini 3.6 FlashGoogle | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Claude Fable 5Anthropic | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Claude Opus 5Anthropic | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Claude Opus 4.8Anthropic | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Claude Opus 4.7Anthropic | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Kimi K3Moonshot AI | 20.3Measured | 77.0Measured | 83.4Measured | 71.0Measured | 74.4Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| DeepSeek V4 ProDeepSeek | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| DeepSeek V4 FlashDeepSeek | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| OpenAI GPT-5.4OpenAI | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| OpenAI GPT-5.2 ThinkingOpenAI | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Claude Opus 4.6Anthropic | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Gemini 3.1 ProGoogle | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Gemini 3 ProGoogle | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| DeepSeek-R1DeepSeek | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Gemini 3.7 FlashGoogle | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| GLM-5.3-FlashZ.ai | 51.6Measured | 65.6Measured | 89.1Measured | 48.4Measured | 76.9Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Claude Fable 5.1Anthropic | 81.3Measured | 96.7Measured | 94.0Measured | 83.9Measured | 75.6Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Gemini 3.8 FlashGoogle | 31.3Measured | 78.7Measured | 76.5Measured | 67.7Measured | 76.9Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| Qwen3.8-FlashAlibaba Cloud | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| GPT-6 AstraOpenAI | 92.2Measured | 90.2Measured | 83.5Measured | 100.0Measured | 87.2Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
| DeepSeek-V3DeepSeek | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Tencent Hunyuan-LargeTencent | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| Hunyuan-T1Tencent | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | —No snapshot score | 0 / 5 available≈ is provisional; unsupported is explicit 0. |
| DeepSeek-V4.1-FlashDeepSeek | 42.2Measured | 63.9Measured | 86.7Measured | 41.9Measured | 88.5Measured | 5 / 5 available≈ is provisional; unsupported is explicit 0. |
Evidence Rules
Real zero scores are preserved. Missing scores stay empty. Overall is the equal-weight average of all five 0–100 capability indices; it requires all five scores and is separate from the AA Intelligence Index.