Claude Code costs up to $44.22 per task on Terminal-Bench 4.1, with max tier off the Pareto line
ArtificialAnlys · x · 2026-11-14
Artificial Analysis updated its FrontierCode 2.0 index using Terminal-Bench 4.1, crediting cognition and the Terminal-Bench team for the underlying benchmarks. Key numbers: Claude Code with Claude Fable 5.2 costs $2.89/task at medium (Index 53), $7.51 at high (58), $15.47 at xhigh (58) and $44.22 at max (56)—with medium, high and xhigh all on the Pareto line. Codex with GPT-6.1 Sol (max) scores 52 at just $1.51 per task. Takeaway: pricier reasoning tiers don't reliably buy higher scores, and the max tier is clearly off the cost-performance frontier.
Related event: Artificial Analysis Coding Agent Index v2.1: Claude Duo Tops Chart(3 posts)→
More from coding & agent
- Free metroidvania level generator: type a seed, get a tiled world exportable as JSON — majidmanzarpour · 2026-11-14
- Replit Agent can now buy for your business, with Visa and Link powering payments — amasad · 2026-11-14
- levelsio's Hotelist adds MCP connector so AI agents can book hotels — alvelda · 2026-11-14
- AWS proposes MixUA: multi-granularity memory scoping for multi-user multi-agent systems — AWS · 2026-11-14
- Apollo Labs launches Tax Law Search MCP with hash-verified citation workflow — Elgon2003 · 2026-11-14
- Developer details why orchestrated coding agents stall mid-backlog and won't chain tasks — W1141175 · 2026-11-14