Anthropic's Sonnet 5.5 Launches at Same Price as Sonnet 5 with Faster Performance

Anthropic announced on September 28 that Sonnet 5.5 was launched at the same base price as Sonnet 5. The company stated the model executes 30% faster and costs up to 30% less per task.
Independent benchmarking firm Artificial Analysis tested the model to its limits and reached different conclusions on cost efficiency.
Sonnet 5.5 is the second model in the Claude 5.5 product line, following Anthropic's release of Opus 5.5 on September 22. On the same day, OpenAI also unveiled GPT-6 Sol and Luna models.
The new model maintains pricing at $2 per million input tokens and $10 per million output tokens, matching Sonnet 5. Anthropic reported Sonnet 5 achieved a 70.6% score on the Terminal-Bench 4.0 agent coding test, while Opus 5.5 reached 66.4%.
Anthropic clarified that Sonnet 5.5 does not expand beyond existing capabilities, focusing instead on preventing misleading use or serious misuse in alignment reviews.
Sonnet 5.5 includes cybersecurity features and a counter-disinformation classifier. It also blocks attempts to extract model reasoning for competitive system training. Claude Haiku 5.5 is expected to launch within weeks.
Meanwhile, OpenAI has delayed the release of its next-generation GPT-6.1 Astra model due to safety concerns, with CNBC reporting it did not meet company standards as of Monday.
Artificial Analysis, which ranked Grok 4.7 fourth in benchmarks, gave Sonnet 5.5 an intelligence index score of 56, placing it second overall—two points below Opus 5.5. The firm reported Sonnet 5.5 scored 64% on Terminal-Bench 4.0 compared to 60% for both Opus 5.5 and GPT-6 Astra in knowledge work tests.
However, the company noted Sonnet 5.5 used significantly more tokens to achieve these results—approximately 60% higher than Opus 5.5 or Sonnet 5 at maximum performance levels, and about seven times more than GPT-6 Astra's peak usage.
Early adopters cited by Anthropic reported opposite outcomes compared to Sonnet 5 in different work environments; Balyasny Asset Management experienced significantly lower token usage for financial tasks. Artificial Analysis tested a pre-release version containing a recently fixed structural output bug and plans to re-evaluate soon.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 앤트로픽 소네트 신모델, 오푸스급 성능…토큰 더 소모