Anthropic releases Claude Sonnet 5.5 with 30% speed and cost gains
Anthropic announced Claude Sonnet 5.5 on September 28, 2026, positioning it as its mid-tier flagship model. The new model is designed for daily tasks, offering output generation speeds over 30% faster than the previous Sonnet 5. While the API unit price remains unchanged at $2 per 1 million input tokens and $10 for output, Anthropic claims the cost to complete the same task is reduced by up to…
Key points
- Claude Sonnet 5.5 is 30% faster and reduces task costs by up to 30% compared to Sonnet 5.
- Anthropic benchmarks show Sonnet 5.5 scoring 1844 on GDPval-AA, nearly matching Opus 5.5's 1846.
- API pricing remains $2 per 1 million input tokens and $10 for output, half the cost of Opus 5.5.
According to benchmarks released by Anthropic and reported by VentureBeat, Sonnet 5.5 performs nearly on par with Opus 5.5 in practical tasks, scoring 1844 on GDPval-AA compared to Opus’s 1846. In coding evaluations, Sonnet 5.5 actually outperforms Opus 5.5, achieving 70.6% on Terminal-Bench 4.0 versus 66.4%. However, VentureBeat notes that these figures are self-reported and may not reflect all production environments. Anthropic recommends using Sonnet for clear, routine tasks like debugging and document creation, while reserving Opus for complex, ambiguous decisions requiring continuous judgment.
For businesses, the combination of speed and cost efficiency offers significant advantages, with client reports citing processing speed increases of up to 2.4 times. Anthropic also plans to release the lighter Haiku 5.5 model within the next few weeks. Users are advised to check the official Claude website for specific plan availability, as it remains unconfirmed whether Sonnet 5.5 is included in the free tier.
Model pages: Claude Sonnet 5.5 → · Claude Opus 5.5 →
The story so far
2 episodes →- Anthropic releases Claude Sonnet 5.5 with 30% speed and cost gainsthis story
Claude Sonnet 5.5 Arrives: The Strategy of Using Sonnet for Daily Tasks and Opus for Complex Decisions
note.com · 4 October 2026
Loading the full article…
This text was published by note.com and written by gswing_taka. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
2sourcesThe headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Reflection may release open-weight model to rival Chinese AI · 1 src
- DW-dev-UE releases Apex-2, a 3.87B MoE LLM trained on 86.5B tokens · 1 src
- SentinelOne engineer uses GPT-6 Astra to crack 217-year-old Napoleonic cipher · 1 src
- Anthropic asks users to share voice data for AI model training · 1 src
- Aleph Alpha releases Kolibri-1 MoE model with 78B parameters and 1M token context · 3 src
Comments
via GitHub Discussions