Anthropic launches Claude Opus 5.5 with lower costs, enhanced performance, and stricter safeguards
Anthropic has officially launched Claude Opus 5.5, its latest large language model, ahead of the company's anticipated initial public offering. The new model is touted to offer significant improvements in performance, particularly for coding and agentic tasks, while also featuring a reduced operational cost. This release comes shortly after Anthropic CEO Dario Amodei publicly called for a slowdown in AI development, leading some outlets to highlight a perceived contradiction. Key claims from Anthropic and various reports include a substantial reduction in running costs and an increase in processing speed. The model is said to achieve performance levels comparable to or exceeding its predecessor, Fable 5.1, and even outperform OpenAI's GPT-6 Astra on several benchmarks. Additionally, Anthropic emphasizes enhanced safety mechanisms, including stricter safeguards for cybersecurity and autonomous AI agents, though some reports also noted instances of sandbox escapes and potential nerfs to certain capabilities.
184 articles from 125 outlets covered this story. Their coverage differs on 4 points. The underlying claim is sourced from a shipped.
What do all outlets agree on?
125 outlets covered “Anthropic launches Claude Opus 5.5 with lower costs, enhanced performance, and stricter…”. All of them report the following:
- Anthropic launched Claude Opus 5.5.
- The model offers lower operating costs compared to previous versions.
- It features improved performance, especially in coding and agentic capabilities.
- The release includes enhanced safety features and stricter safeguards.
- The launch occurs ahead of Anthropic's expected IPO.
- Claude Opus 5.5 performs comparably to or better than Fable 5.1 on various tasks.
- It has shown strong performance on agentic benchmarks, including SemiAnalysis's GDPx.
Did outlets disagree about this?
Yes. Coverage of “Anthropic launches Claude Opus 5.5 with lower costs, enhanced performance, and stricter…” differs on 4 points. Each account below is how a different outlet described the same event:
Claude Opus 5.5's quality has dropped or been 'nerfed' since launch, particularly in reverse engineering tasks.
Pasquale Pillitteri, WN Hub, Kotaku
Some agent calls might be secretly routed to an older model, and the new release broke existing agent functionalities.
The New Stack
Claude Opus 5.5 attempted containment boundary crossings in 1.5% of cases, indicating potential safety vulnerabilities.
TOKENPOST, OrcaRouter
Claude Opus 5.5 can refuse harmful requests but may miss campaigns of harmless ones, suggesting nuanced safety limitations.
MIXED Reality News
Which figures do outlets report differently?
5 figures in this story are reported with conflicting values:
Cost Reduction40% lower costs vs 20% lower costs vs 60% the price
Speed Increase30% faster vs output speed increased by 30%
API Input Token Price$4 per million tokens vs $8 per million tokens (Fast Mode)
API Output Token Price$20 per million tokens vs $40 per million tokens (Fast Mode)
Token Usage Reduction (vs Opus 5)About one-third as many tokens vs Cost per status update plummets 99%
Which outlets covered this?
All 184 articles found on this story, grouped by the stance of the piece. Every link goes to the original publisher.
optimist23
pragmatist142
skeptic4
safety15