Coding agentArtificial AnalysisConfirmed
Story 13 of 15 in the Friday 13 November editionArtificial Analysis ships Coding Agent Index v2.1 with FrontierCode
Released on 14 November 2026, v2.1 upgrades to 132-task FrontierCode-AA 2.0 and reward-hack-hardened Terminal-Bench 4.1, with Claude Fable 5.2 and Claude Opus 5.5 tied at 58.
Edited by Ashutosh Shrivastava
14 Nov 2026, 09:50 IST · 2 min read
Artificial Analysis launched Coding Agent Index v2.1 on 14 November 2026, upgrading to 132-task FrontierCode-AA 2.0 and verifier-hardened 66-task Terminal-Bench 4.1 alongside SWE-Atlas-QnA-AA. Average scores fell 5 points; Claude Fable 5.2 (high/xhigh in Claude Code) and Claude Opus 5.5 (max in Claude Code or with GPT-6 Luna High in Devin Fusion CLI) tie at 58, while task costs above 48 span $1.51 to $44.22.
Technical details
- Benchmark suite composition
- FrontierCode-AA 2.0 (132 tasks), Terminal-Bench 4.1 (66 tasks) and SWE-Atlas-QnA-AA
- Top leaderboard scores
- 58 (Claude Fable 5.2 high/xhigh; Opus 5.5 max), 56 (Sonnet 5.5 max; Fable 5.2 max), 55 (Gemini 4 Carbon high)
- Cost-per-task extremes (48+)
- $1.51 for GPT-6.1 Sol (max, score 52) in Codex up to $44.22 for Claude Fable 5.2 (max, score 56) in Claude Code
Release details & context
Timeout penalties at maximum thinking and GPT-6.1 Terra score stability
Artificial Analysis highlighted two harness-level effects in v2.1: raising Claude Fable 5.2 from high or xhigh up to max effort inside Claude Code actually dropped its composite score from 58 to 56 because longer reasoning runs hit task time limits more frequently.
Meanwhile, OpenAI’s Codex paired with GPT-6.1 Terra (max) was the sole evaluated configuration that avoided the 5-point average drop caused by FrontierCode-AA 2.0, holding within 1 point of its v2.0 mark, while Gemini 4 Carbon (high) scored 55 inside Antigravity CLI.
Sources
Artificial Analysis release timeline (3 launches tracked)
- Artificial Analysis launches Coding Agent Index v2.0 across 290 tasks
Artificial Analysis released Coding Agent Index v2.0 across 290 tasks; Claude Opus 5.5 in Claude Code ranks first at 61.7, ahead of Sonnet 5.5 at 59.3.
- Artificial Analysis updates AA-Omniscience v1.1 knowledge and hallucination benchmark
Artificial Analysis launched AA-Omniscience v1.1 (3k questions), led by Gemini 4 Carbon (High) at 54 Net Score and Gemini 4 Argon at 4% hallucination.