Anthropic shipped Claude Opus 5.5 on 22 September 2026, and the headline isn’t a capability leap so much as a cost collapse: input tokens now run $4 per million and output $20 per million, a 40% drop versus Opus 5, with cached reads down 60% to 20 cents. Output generation is also over 30% faster. On Anthropic’s own benchmarks, Opus 5.5 matches or beats its flagship Fable 5.1 model on agentic coding, computer use, visual chart recognition and multidisciplinary reasoning — meaning the model previously priced as a premium tier now undercuts it while performing at a comparable level for most coding work.
A frontier tier is being priced like a mid tier
This is the third meaningful repricing of Anthropic’s coding-capable models in under three months, following the Sonnet 5 permanent price freeze in August and the Opus 5 launch discount in July. The pattern holds across the industry: OpenAI answered Opus 5.5’s launch the same day with two cheaper GPT-6 variants. For teams running agentic coding workloads — Claude Code chief among them — the frontier tier that used to carry a clear cost premium over mid-tier models is now priced close enough to erode the reason to default to a cheaper model for production coding work.
So what
If you’re commissioning a build and a development partner’s cost estimate still assumes last quarter’s token pricing, ask them to re-run it — the gap between “frontier model” and “affordable model” has narrowed enough that it changes which model makes sense as the default for a given project, not just the total bill. It also means AI-assisted development is now viable for a wider range of project budgets than it was even two months ago. We treat model selection and pricing as a live input to project scoping rather than a fixed assumption — see how that shapes our AI-assisted development approach, or get in touch to scope a build against current model economics.