Breaking

Anthropic Launches Claude Opus 5.5, Claiming a CursorBench Coding Edge at About One-Third GPT-5.6 Sol Task Cost

September 23, 2026 at 10:03 EDT

Foundation Models
AI Agents
Software Dev & Coding

Anthropic introduced Claude Opus 5.5 on Sept. 22, 2026, positioning the new Opus model for long-running coding agents and knowledge work. The company says the model beats GPT-5.6 Sol on CursorBench 4.0 at roughly one-third the per-task cost, while keeping a 1M-token context window and 128K-token maximum output.

Sept. 22, 2026 · Anthropic

Opus 5.5 targets coding-agent economics

Anthropic says its new Opus model beats GPT-5.6 Sol on CursorBench while running the task at about one-third the cost.

52.5%

Opus 5.5 default-effort CursorBench score

41.7%

GPT-5.6 Sol top CursorBench score

about 1/3

Opus 5.5 task cost versus Sol

Token context window

CursorBench 4.0 score

52.5%

Opus 5.5

41.7%

GPT-5.6 Sol

Relative task cost on CursorBench

about 1x

Opus 5.5

about 3x

GPT-5.6 Sol

The task-cost claim reflects fewer tokens and effort settings, not a lower standard input/output token rate.

Developer upside

Early reports emphasize long unattended coding runs, fewer steps and lower token use on large migrations, audits and reviews.

Integration caveats

Teams must recheck effort settings, forced tool-use patterns, thinking blocks and safeguard-driven refusals before switching production workloads.

Continue reading

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