Breaking

Anthropic Launches Claude Opus 5.5 as Lower-Cost Agentic Coding Model Draws Game-Building Tests

September 22, 2026 at 22:45 EDT

Foundation Models
AI Agents
Software Dev & Coding

Anthropic has launched Claude Opus 5.5, a new frontier model aimed at long-running coding agents, research and professional knowledge work, with lower pricing and stronger efficiency claims than Opus 5. The release quickly drew attention from developers testing whether the model can turn broad prompts into playable software, including game prototypes.

September 22, 2026 · Anthropic

Claude Opus 5.5 shifts the agent race toward cost per task

Anthropic's new model targets long-running coding agents and knowledge work, with lower typical costs, faster output and early game-building tests that show both promise and budget risk.

Typical workload cost

40%

Less than Opus 5, according to Anthropic.

Terminal-Bench 4.0

66.4%

Reported agentic terminal coding score.

Context window

Tokens for large projects and long sessions.

Typical cost moves from 100 to 60

100

Opus 5 baseline

Opus 5.5 typical cost

Each block represents 10 units of the Opus 5 baseline cost.

What testers like

Clearer explanations, stronger instruction-following and early playable game demos from broad prompts.

What still needs control

Long autonomous runs can consume usage, and self-review judgments may not be reliable without a clear rubric.

Agent workflow now needs explicit stopping rules

Prompt and context

Tools and subagents

Test and self-review

Budget and stop point

The strongest demos come from letting the model work, but production use depends on defining what done and enough mean.

Continue reading

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