Breaking

Anthropic Debuts Claude Opus 5.5, Cutting Costs While Lifting Agent Benchmarks

September 22, 2026 at 12:37 EDT

Foundation Models
AI Agents
Software Dev & Coding

Anthropic released Claude Opus 5.5 on Sept. 22, 2026, positioning it as the first model in its Claude 5.5 family and saying it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5. The launch gives Anthropic a lower-cost flagship for coding agents and knowledge work while extending the same class of safeguards it uses for its most capable models.

Sept. 22, 2026 - Anthropic

Claude Opus 5.5 lowers the cost of frontier agents

Anthropic says its first Claude 5.5 model reaches Fable 5.1-level work on most tasks, improves coding and knowledge benchmarks, and runs typical workloads for 40% less than Opus 5.

Typical workload cost

40%

less than Opus 5

API token prices

$4/$20

per 1M input/output

Terminal-Bench 4.0

66.4%

agentic coding score

Output generation

30%+

faster than Opus 5

Typical workload cost index

Ten blocks represent Opus 5 cost. Six blocks represent Opus 5.5 after the reported 40% reduction.

Opus 5

100 index

Opus 5.5

60 index

The price story is proportional, not just a headline discount.

Lower per-token pricing and fewer tokens per task combine to make long-running agent sessions cheaper.

Safeguard path for high-risk work

User request

Safety classifier

Allowed work

Fallback or verified access

Cybersecurity, biology and distillation safeguards remain part of the product design, even as the model becomes more capable.

What users gain

Early tests emphasize fewer steps, lower token use, clearer writing and stronger long-session coding performance.

What still needs proof

Benchmark margins may not fully predict deployment behavior, and safeguard fallbacks can shape results in high-risk domains.

Continue reading

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