Stay informed with free updates
Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.
Anthropic and OpenAI both released cheaper versions of their leading models on Tuesday, as the price war between the two leading US labs and their low-cost Chinese rivals intensifies.
The latest Claude model from Anthropic, Opus 5.5, is roughly 40 per cent cheaper to use than its precursor and was the first of a new “family” of models that will be more cost-efficient than previous generations, the company said.
OpenAI, meanwhile, released lower-cost versions of ChatGPT-6, known as Sol and Luna, following the release of its most advanced ChatGPT-6 Astra model at the start of the month. It said the models will be 50 per cent cheaper than their predecessors in the ChatGPT-5.6 slate.
The rival US AI labs are increasingly competing on both price and performance for cost-conscious business customers, after a period characterised by an all-out race for the highest capabilities.
Anthropic is aiming to bolster its customer base ahead of a planned IPO this autumn that is expected to be the largest of all time. Investors have speculated that a new model release would be the catalyst for the company to release its long-awaited IPO prospectus.
They are closely watching for signs that the group can sustain its blistering growth rate in the coming years amid intensifying competition and increasingly price-sensitive customers.
OpenAI has said it is unlikely to go public before 2027. It has slashed the prices of its models in recent months, seeking to kick-start its growth after a sluggish start to the year. The company also claims its Astra model is the most powerful model on the market.
Anthropic’s Opus 5.5 will cost $4 per million input tokens and $20 per million output tokens. Tokens are the units of data that comprise AI queries. Opus would be able to perform most tasks at the same level as the group’s most powerful tool, Fable 5.1, the company said.
Fable 5.1 is priced at $10 per million input tokens and $50 per million output tokens, the same as OpenAI’s Astra.
Its next-most-powerful Sol model costs $2 per million input tokens and $10 per million output tokens, while the lightweight Luna version costs $0.10 per million input tokens and $0.50 per million output tokens.
Both Anthropic and OpenAI are facing renewed threats from “open-weight” models, which users can customise and run on their own hardware. These models, mostly from China, are taking share from US labs, according to data from cloud application company Vercel.
Opus 5.5 is the first model released by the company since its chief executive Dario Amodei called on AI labs to “pace the frontier” of development and more deeply embed third-party testers, in an effort to mitigate AI risks and security breaches after researchers warned that the unchecked race to build self-improving superintelligence could destroy humanity.
Anthropic said its new model had undergone “extensive alignment testing, pre-release evaluation by outside organisations, and [included] safeguards for high-risk areas like cyber security and biology”.
OpenAI on Tuesday set its priorities for further work with independent evaluators — including training, safeguards, preparedness and investigating incidents — after Anthropic last week said it would work with Accenture on safety testing.