Claude Sonnet 5.5: what it already changes for custom software

On September 28, 2026, Anthropic shipped Claude Sonnet 5.5, the second model in its 5.5 family after Opus 5.5 (Anthropic; also confirmed by TechCrunch). It lands six days after we covered how Opus 5.5 and GPT-6 Sol/Luna kicked off the tiered AI era, and it turns that prediction into fact: Anthropic’s mid tier, built for the bulk of everyday development work, is now actually available. At Evicron, an AI and custom software studio based in Barcelona, what matters isn’t the launch itself, but which tasks in a real project it can already handle without touching the top tier.

What Sonnet 5.5 actually brings

Sonnet 5.5 keeps the same per-token price as its predecessor — $2 per million input tokens and $10 for output, with cached reads at $0.20 and cache writes at $2.50 — half of Opus 5.5’s price. But it does more for that price: it’s over 30% faster than Sonnet 5 and up to 30% cheaper per completed task, because it needs fewer steps and iterations to reach the same result. On Anthropic’s own benchmarks it jumps from 10.3% to 70.6% on Terminal-Bench 4.0, and on the GDPval-AA v2.1 knowledge-work test it scores 1,844 points against Opus 5.5’s 1,846 — essentially tied. It’s already available as claude-sonnet-5-5 on the Claude API, AWS, Google Cloud and Microsoft Azure.

The three-tier lineup, almost complete

When we covered the Opus 5.5 launch, Anthropic had already said Sonnet 5.5 and Haiku 5.5 were coming “in the coming weeks” to round out the lineup. Sonnet 5.5 delivers on that: it’s the fast, affordable complement for well-scoped tasks, while Opus 5.5 stays the model to reserve for complex, sustained reasoning — reviewing long contracts, decisions with real financial stakes. Haiku 5.5, the budget tier, still hasn’t arrived, so the cheapest slot for high, predictable volume — triaging tickets, pulling fields from a form — is still waiting to be filled.

What it can already handle in your software

Anthropic and several enterprise customers have published concrete figures on this tier shift:

• Bug fixing and routine development — the task where Sonnet 5.5 performs best per dollar, versus the open-ended reasoning that’s still Opus territory.
• Polished documents, slides and spreadsheets, and UI tweaks, where speed matters more than deep reasoning.
• Zendesk reports support tickets processed 20% faster; Box cites processing 2.4x faster; Atlassian says its agents run up to 30% faster on the new model.

None of these cases replace Opus where sustained reasoning genuinely matters; what changes is that there’s now a real mid tier, not just a “cheaper” model with worse output.

What it means for a custom software project

1. Check what you’re routing to Opus by default. If a feature in your software does routine work — reviewing code, drafting answers from internal documentation, generating a report — it probably fits better, and cheaper, on Sonnet 5.5.
2. Don’t switch models without measuring before and after. The real gain is output per euro spent, not chasing the newest release; compare quality and cost on your own use cases before moving production traffic.
3. Leave Haiku’s job for when it arrives. Forcing today’s mid tier onto very high-volume, low-complexity work can still cost more than waiting a few more weeks for the budget tier.
4. Architecture matters more than the model of the moment. If your integration talks to the provider through a middle layer, adopting Sonnet 5.5 for the tasks that fit it is a config change, not a new project.

How we handle this at Evicron

In our applied AI projects, we audit which tasks in your software are running on a more expensive tier than they need, and migrate the right ones to Sonnet 5.5 without touching the user experience. If you’re wondering whether your custom software splits work well across AI tiers, our AI consulting for businesses gives you that answer with numbers, not guesswork.

Bottom line

Sonnet 5.5 confirms Anthropic’s three-tier lineup is no longer a promise: the mid tier is here, performs nearly on par with Opus on knowledge work, and costs half as much. The opportunity for any software with AI built in is to check, task by task, whether it’s still overpaying for the wrong tier.

Want to know if your custom software is using the right Claude tier for each task? Get in touch: the first consultation is free and we reply within 24 hours.