Breaking

Claude Opus 5.5 High Reaches No. 2 in Agent Arena at Lower Median Task Cost

September 28, 2026 at 13:02 EDT

Foundation Models
AI Agents
Software Dev & Coding

Claude Opus 5.5 (High) has entered Agent Arena at No. 2 with a +12.15% net improvement score, trailing only Claude Fable 5.1 (Max) while posting a much lower median task cost. The benchmark result strengthens Anthropic's case that its newest Opus model is not only a higher-capability agentic system but also a more cost-efficient option for long-running AI work.

Sep. 28, 2026 · Anthropic

Opus 5.5 pushes the agent cost frontier

Claude Opus 5.5 (High) entered Agent Arena at No. 2 with a +12.15% net improvement score, while its $1.31 median task price undercuts Claude Fable 5.1 (Max) by about 64%.

No. 2

Agent Arena rank

+12.15%

Net improvement

$1.31

Median task price

64%

Lower than Fable 5.1 Max

Median task cost, shown to scale

The top-ranked Fable configuration costs more than three times as much per median task.

$1.31

Claude Opus 5.5 (High)

$4.14

Claude Fable 5.1 (Max)

Why teams like it

Developers report stronger autonomous coding, better long-context collaboration and lower token use for agent workflows.

What needs care

Existing Opus 5 integrations may need fresh effort sweeps, thinking-state checks and computer-use migration work.

The operational tradeoff

Higher agent score→Lower median task cost→Migration checks required

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.