Anthropic shipped Claude Sonnet 5.5 this week: 30% faster, 30% cheaper on most tasks, and pricing unchanged from the previous generation—a clear signal that it's going head-to-head with OpenAI for end users. What caught our editorial attention is that this mid-tier model has replaced the underlying engine of the claude.ai free tier, and now outperforms ChatGPT's free tier (which runs on Luna 5.6).
What this is
Claude is Anthropic's flagship large-model product (think ChatGPT equivalent), split by capability into three tiers: Haiku (lightweight), Sonnet (mid-tier), and Opus (flagship). Sonnet 5.5 is the new middle-tier release.
Anthropic claims more than 30% speed improvement and 30% cost reduction on most tasks at unchanged pricing—meaning Anthropic is absorbing part of the inference (the compute that runs the model) cost itself. On several coding benchmarks, Sonnet 5.5 approaches the level of the previous-generation flagship Opus 5.5.
There are flaws. Prominent blogger Simon Willison found during testing that 5.5 inherits a bug from Opus 5.5: at "max" reasoning effort, the model chewed through 128,000 tokens (the smallest unit of text a model processes) without completing the task, then failed after burning $1.28.
Industry view
The dominant read is that Anthropic is waging a price war: by giving the mid-tier model away free, it forces OpenAI to respond, and the free-tier capability gap will naturally pull some light users over.
There's pushback. One view: Sonnet 5.5's lower cost mainly reflects efficiency gains in Anthropic's inference infrastructure—not a sign that overall large-model costs are falling. OpenAI is walking a different path—more complex multi-step reasoning tasks, where the pricing curve may not move in lockstep. Another worry is whether the "give-it-away-to-grab-users" strategy is sustainable: Anthropic is still bleeding cash, and its fundraising cadence determines how long it can keep burning. If the next funding round doesn't close on favorable terms, the free tier's financial footing thins out overnight.
Impact on regular people
For enterprise IT: Teams running internal tools on the Claude API (paid per-call access) should pilot Sonnet 5.5 on small traffic before considering a full switch—don't migrate in one shot. 5.5 still carries the 5.5-line's reasoning bug.
For working professionals: For emails, document edits, and summaries, the free Claude tier is now almost certainly good enough—no need to keep paying for ChatGPT Plus.
For the consumer market: The underlying models powering mobile AI assistants and writing apps will reshuffle—who's willing to give up margin and who sticks with paywalls will be decided in the next six months.