Efficient frontier
Models on the Pareto frontier by API cost per task and DeepSWE v1.1.
Coding agent comparisons
Each point is one model running in one agent harness at one effort setting, plotted by benchmark score against cost, time, or tokens per task.
The Coding Agent Index v1.5 Artificial Analysis coding-agents snapshot contains 31 configurations across 22 models and 9 agent harnesses. Retrieved Oct 5, 2026, 2:10 PM UTC.
Loading chart…
All 31 configurations
| Model (setting) | Agent | Provider | AA Index | DeepSWE v1.1 | Terminal-Bench 4 | SWE-Atlas-QnA | Cost per task | Time per task | Tokens per task |
|---|---|---|---|---|---|---|---|---|---|
| Sonnet 5.5 (max) | Claude Code | Anthropic | 68.4 | 72.0 | 66.2 | 66.9 | $14.19 | 87m | 27.7M |
| Opus 5.5 (max) | Claude Code | Anthropic | 66.0 | 68.4 | 63.1 | 66.4 | $13.04 | 64m | 15.6M |
| Gemini 4 Argon (default) | Antigravity CLI | 63.8 | 78.8 | 56.1 | 56.5 | $5.84 | 35m | 13.7M | |
| GPT-6.1 Sol (xhigh) | Codex | OpenAI | 62.9 | 73.2 | 54.5 | 61.0 | $1.04 | 16m | 3.2M |
| Sonnet 5.5 (xhigh) | Claude Code | Anthropic | 62.9 | 68.4 | 58.1 | 62.1 | $3.33 | 27m | 7.1M |
| Fable 5.1 (with fallback) (max) | Claude Code | Anthropic | 62.2 | 64.3 | 57.6 | 64.8 | $12.39 | 35m | 5.7M |
| Claude Fable 5.1 XHigh + SWE-2 Medium (default) | Devin Fusion CLI | Cognition | 61.7 | 63.1 | 56.1 | 65.9 | $7.90 | 36m | 9.7M |
| GPT-6 Astra (max) | Codex | OpenAI | 61.6 | 67.6 | 55.6 | 61.8 | $7.47 | 29m | 3.3M |
| GPT-6.1 Sol (medium) | Codex | OpenAI | 61.4 | 72.0 | 51.5 | 60.8 | $0.705 | 11m | 2.3M |
| GPT-6.1 Sol (high) | Codex | OpenAI | 60.1 | 70.5 | 50.0 | 59.9 | $0.889 | 13m | 2.8M |
| GPT-6.1 Sol (max) | Codex | OpenAI | 60.1 | 69.6 | 53.0 | 57.8 | $1.55 | 24m | 4M |
| Opus 5 (max) | Claude Code | Anthropic | 59.7 | 62.5 | 54.5 | 62.1 | $10.79 | 42m | 11.4M |
| GPT-6 Astra XHigh + SWE-2 Medium (default) | Devin Fusion CLI | Cognition | 58.9 | 67.3 | 50.0 | 59.4 | $4.54 | 25m | 6.1M |
| GPT-6.1 Sol (low) | Codex | OpenAI | 57.2 | 67.6 | 49.0 | 55.1 | $0.499 | 8.6m | 1.7M |
| GPT-6 Sol (max) | Codex | OpenAI | 56.7 | 69.0 | 43.4 | 57.5 | $2.99 | 22m | 9.8M |
| Grok 4.7 (xhigh) | Grok Build | xAI | 56.3 | 72.6 | 33.3 | 62.9 | $8.82 | 39m | 14.3M |
| Sonnet 5.5 (high) | Claude Code | Anthropic | 55.0 | 66.7 | 41.9 | 56.5 | $1.24 | 12m | 2.7M |
| GPT-5.6 Sol (max) | Codex | OpenAI | 54.6 | 72.3 | 37.4 | 54.0 | $6.35 | 21m | 10.2M |
| Muse Spark 1.3 (max) | Muse Code | Meta | 54.3 | 71.7 | 31.8 | 59.4 | $3.98 | 18m | 16.7M |
| GLM-5.3 (default) | Opencode | Z.ai | 53.6 | 61.4 | 39.9 | 59.4 | $4.24 | 48m | 14.3M |
| Kimi K3 (default) | Kimi Code CLI | Moonshot AI | 51.9 | 68.4 | 21.2 | 66.1 | $5.05 | 61m | 8.5M |
| Muse Spark 1.3 (xhigh) | Muse Code | Meta | 48.3 | 73.2 | 17.2 | 54.6 | $3.47 | 18m | 16.2M |
| Grok 4.6 (xhigh) | Grok Build | xAI | 47.0 | 64.9 | 17.7 | 58.3 | $3.57 | 19m | 5.5M |
| Sonnet 5.5 (medium) | Claude Code | Anthropic | 45.9 | 65.5 | 27.3 | 44.9 | $0.619 | 8.5m | 1.3M |
| Qwen3.8 Max (default) | Claude Code | Alibaba Cloud | 43.3 | 51.0 | 16.7 | 62.1 | $3.48 | 63m | 8.7M |
| GPT-5.6 Luna (max) | Codex | OpenAI | 43.2 | 66.4 | 14.6 | 48.7 | $0.438 | 24m | 14.9M |
| DeepSeek V4 Pro 0813 (max) | Codex | DeepSeek | 43.1 | 57.2 | 10.1 | 61.8 | $0.238 | 40m | 24.2M |
| Sonnet 5.5 (low) | Claude Code | Anthropic | 42.1 | 61.9 | 25.3 | 39.0 | $0.483 | 6.3m | 977.3K |
| Gemini 3.8 Flash (high) | Antigravity SDK | 41.9 | 65.8 | 14.6 | 45.2 | $2.47 | 12m | 13.5M | |
| GPT-6 Luna (max) | Codex | OpenAI | 41.1 | 63.7 | 15.2 | 44.4 | $0.176 | 21m | 10.2M |
| DeepSeek V4 Flash 0731 (max) | Codex | DeepSeek | 38.7 | 54.3 | 10.6 | 51.3 | $0.085 | 18m | 15.4M |
