ConstellationAI Model Pulse

Board · Repo work

Coding agents

Artificial Analysis Coding Agent Index — average of DeepSWE, Terminal-Bench v2, and SWE-Atlas-QnA. This ranks agent + model + settings, not a raw base model.

Repo work · Index %

Coding agents

Updated 2026-09

RankModelIndex %Notes
01Claude Code · Fable 5.1 maxAnthropic70.4With fallback
02Claude Code · Opus 5 xhighAnthropic68.1Adaptive reasoning
03Muse Code · Spark 1.3 maxMeta68.0Agent harness
04Claude Code · Fable 5 maxAnthropic67.2With fallback
05Claude Code · Opus 5 maxAnthropic67.0Tied with Codex Astra
06Codex · GPT-6 Astra maxOpenAI67.0Tied with Opus 5 max
07Claude Code · Fable 5 xhighAnthropic66.0With fallback
08Codex · GPT-5.6 Sol maxOpenAI65.1Sol variant

Artificial Analysis Coding Agent Index — average of DeepSWE, Terminal-Bench v2, and SWE-Atlas-QnA. This ranks agent + model + settings, not a raw base model.

Artificial Analysis coding agents