Claude Sonnet 5 Is Live — Anthropic's New Workhorse for Coding and Agents

July 1, 2026

Claude Sonnet 5 is in the model picker today. It's Anthropic's newest Sonnet — near-Opus intelligence for coding and agentic workflows, a 1M-token context window, always-on adaptive thinking, and 128K max output. It sits in the Premium pricing band (same per-token rates as Sonnet 4.6), and the Auto router can pick it in the Workhorse tier when you're not pinned to a specific model.

What you can do

• Pick Claude Sonnet 5 — Reasoning from the model selector on any plan.
• Long-horizon coding and agent work — multi-step refactors, tool-heavy research loops, and projects where Sonnet 4.6 was good but you want the latest Anthropic workhorse.
• 1M context, adaptive thinking — the model decides how deeply to reason per turn; effort defaults to medium in the catalog.
• Swarm-eligible — multi-agent plans can assign Sonnet 5 to worker steps when EU Data Residency is off.
• Auto-routable — tagged Workhorse tier with a higher quality score than Sonnet 4.6, so Auto may prefer it on complex turns.

Where this shows up

• You're running a long agent loop with many tool calls and want Anthropic's newest Sonnet without jumping to Opus pricing.
• You have a large codebase or document corpus that fits in one pass and need strong coding quality at Premium-band cost.
• Sonnet 4.6 was your default Anthropic pick — Sonnet 5 is the direct upgrade in the same band.

Try it

• "Refactor this module for async I/O, keep the public API stable, and explain every breaking-risk change."
• "Plan a three-step research swarm: market scan, competitor pricing, then a one-page recommendation."
• "Hold this entire repo context and find the three architectural decisions that will hurt us at 10× scale."

Heads up

• Premium band — 6× tokens. Same per-token API cost as Sonnet 4.6 ($3 / $15 per million input/output); billed at the Premium multiplier on your plan quota.
• Global Bedrock routing only. AWS publishes Sonnet 5 with a Global inference profile (global.anthropic.claude-sonnet-5), not an EU geo profile. Requests may route outside the EU for throughput.
• Hidden when EU Data Residency is on. If you've enabled EU Data Residency (Settings → Inference on desktop, Digital Twin maker, or via chat), Sonnet 5 is excluded from the picker, Auto, and Swarm — same rule as other Global-only models. Users in Europe who have not turned that on can still select it.
• Desktop Inference settings unchanged. This release adds the cloud model only. The Inference settings tab (local hardware, model source, EU toggle on desktop) remains Redbox/desktop-only and does not ship to the web app at app.alfrada.ai. Production must keep REDBOX_MODE=false.