Image to help understand the article
Anthropic cuts prices for its latest AI model
Anthropic unveiled Claude Opus 5.5 on Sept. 22, introducing the first model in its 5.5 series with lower operating costs and faster output than its predecessor.
The company said Opus 5.5 delivers performance comparable to Claude Fable 5.1 on most tasks while costing 40% less to run than Opus 5. It also generates output more than 30% faster, according to Anthropic.
API access costs $4 per million input tokens and $20 per million output tokens, down from $5 and $25, respectively, for Opus 5. Tokens are the units AI companies use to measure the text processed and generated by their models.
Claude Opus 5.5 is available through Anthropic’s Claude platform and major U.S. cloud providers Amazon Web Services, Google Cloud and Microsoft Azure. Its API model identifier is claude-opus-5-5.
Mixed results against OpenAI’s Astra
Anthropic’s published FrontierCode v1.1 results gave Opus 5.5 a score of 54.4%, narrowly above the 53.3% reported for OpenAI’s GPT-6 Astra. Anthropic said its model surpassed Astra’s best score using the default medium-effort setting while costing about one-fifth as much per task.
On Terminal-Bench 4.0, Anthropic said Opus 5.5 matched Astra at roughly 40% of the cost. The company’s table showed Opus 5.5 scoring 66.4% at its highest effort setting, compared with Astra’s 57.9% at a high setting.
Astra led in other tests. It scored 64.6% on Terminal-Bench-Science 0.1, compared with 58.7% for Opus 5.5, and posted 41.4% on AutomationBench against Anthropic’s 40.0%. On CursorBench 4.0, Opus 5.5 scored 57.8%, while GPT-5.6 Sol scored 41.7%; no Astra result was provided.
The figures were reported by the companies and have not been independently reproduced. Independent evaluator Vals AI described the two models as nearly even, with Astra slightly ahead. Digital Applied advised treating differences of less than 5 points as inconclusive without direct testing, saying Opus 5.5 generally led on price and shared benchmarks while Astra retained an advantage in automation and science tasks.
Safety gains come with acknowledged limits
Anthropic said Opus 5.5 achieved its highest score to date in an internal audit of autonomous behavior. Attempts to operate outside permitted boundaries fell by about 85% compared with Opus 5, while Opus 5.5 and Fable 5.1 recorded the company’s lowest prompt-injection success rates. Cybersecurity and biology-related requests are subject to safeguards that route work to other models.
The company also acknowledged that safety testing remains imperfect, saying that building evaluations capable of reliably detecting every failure before deployment is still an unsolved problem. Anthropic said it also observed signs that Opus 5.5 often suspects when it is being evaluated, a behavior that can complicate efforts to determine how a model will perform under ordinary conditions.
Release follows CEO’s call to pace AI development
The launch is Anthropic’s first model release since CEO Dario Amodei called for managing the pace of advances in AI capabilities so that risk-prevention efforts have time to keep up.
Opus 5.5 arrives amid continued competition among U.S. artificial intelligence developers. OpenAI offered GPT-6 Astra in a limited release on Sept. 3, while Anthropic announced Claude Fable 5.1 and Claude Mythos 5.1 earlier in the month.
Comments
Post a Comment