Efficient frontier
Models on the Pareto frontier by API cost per task and AA Index.
AA Index — Overall performance across code changes, terminal work, and repository understanding.
Provider ranges
Minimum, median, and maximum AA Index by provider.
Model updates
Daily snapshot diff · checked
- Benchmark change
Qwen3.8 Max
Claude Code · Alibaba Cloud · default
- AA Index
- 58.756.6
- SWE-Atlas-QnA
- 44.938.6
- Benchmark change
Opus 4.7
Opencode · Anthropic · medium
- SWE-Atlas-QnA
- 34.936.0
- New model
Qwen3.8 Max
Claude Code · Alibaba Cloud · default
- AA Index
- 58.7
- DeepSWE
- 51.9
- Terminal-Bench v2
- 79.4
- SWE-Atlas-QnA
- 44.9
- New model
DeepSeek V4 Flash
Codex · DeepSeek · max
- AA Index
- 55.5
- DeepSWE
- 42.8
- Terminal-Bench v2
- 84.9
- SWE-Atlas-QnA
- 38.7
- New model
Gemini 3.6 Flash
Opencode · Google · high
- AA Index
- 45.6
- DeepSWE
- 41.3
- Terminal-Bench v2
- 73.8
- SWE-Atlas-QnA
- 21.8
- Benchmark change
GPT-5.4
Cursor CLI · OpenAI · medium
- Terminal-Bench v2
- 65.566.3
- New setting
Opus 4.8
Claude Code · Anthropic · xhigh · 3 settings
- AA Index
- 58.5
- DeepSWE
- 51.3
- Terminal-Bench v2
- 81.3
- SWE-Atlas-QnA
- 42.7
- New model
Opus 5
Claude Code · Anthropic · xhigh · 5 settings
- AA Index
- 66.7
- DeepSWE
- 60.5
- Terminal-Bench v2
- 84.9
- SWE-Atlas-QnA
- 54.8