Anthropic shipped Claude Sonnet 5 on June 30, 2026, and it immediately became the new default model for Free and Pro plans. The pitch is simple: performance close to Opus 4.8, at a fraction of the cost, with agentic behavior (planning, tool use, autonomous multi-step work) that previously required the larger model.
If you're deciding whether to migrate a workflow, an agent, or an API integration to Sonnet 5, here's what actually changed and how to think about it.
What Changed With Sonnet 5
Sonnet 5 replaces Sonnet 4.6 as the default model across Claude.ai Free and Pro. The headline improvements Anthropic is emphasizing:
- Near-Opus agentic performance — planning, using tools like browsers and terminals, and running autonomously at a level that previously needed Opus-class models.
- Lower cost than Opus 4.8 — and priced below competing frontier models from OpenAI (GPT-5.5) and Google (Gemini 3.1 Pro) at launch.
- Default model status — new conversations on Free and Pro plans route to Sonnet 5 automatically; Opus 4.8 remains available for tasks that need the top tier of reasoning.
The practical read: Sonnet 5 is built for the agentic workloads that were quietly running up enterprise API bills on Opus — long tool-use chains, multi-step coding tasks, and autonomous agents — without the Opus price tag.
Claude Sonnet 5 Pricing
Sonnet 5 launched with introductory API pricing that steps up after a fixed window:
| Period | Input (per million tokens) | Output (per million tokens) |
|---|---|---|
| Launch – Aug 31, 2026 | $2 | $10 |
| After Aug 31, 2026 | $3 | $15 |
Both the intro and post-August rates sit below Opus 4.8 pricing, which is why Sonnet 5 is positioned as the default choice for cost-sensitive agentic work rather than a straight Opus replacement. If you're running high-volume agents or batch jobs, locking in usage before September 1 captures the lower rate — budget for the step-up afterward rather than being surprised by it.
Sonnet 5 vs Opus 4.8 vs Sonnet 4.6
Three models, three jobs. Here's the practical split:
- Sonnet 5 — default choice for most agentic and coding work. Near-Opus capability on planning and tool use, at Sonnet pricing. Start here unless you have a specific reason not to.
- Opus 4.8 — reserve for tasks where you need the absolute ceiling on reasoning depth: complex architectural decisions, ambiguous multi-constraint problems, or anything where a wrong answer is expensive and Sonnet 5's near-Opus performance isn't quite Opus performance.
- Sonnet 4.6 — still available via the API for workloads already tuned and validated against it. There's rarely a reason to start something new on 4.6 now that 5 is out at a comparable or lower price, but don't rip out a working production integration just because a new number shipped.
For a deeper breakdown of when Opus-tier reasoning is worth the cost, see our Sonnet vs Opus guide.
Should You Migrate?
If you're running production workloads on Sonnet 4.6, migrating to Sonnet 5 is low-risk and likely worth it: better agentic performance at similar or lower cost. Standard rollout advice applies:
- Test against your eval suite first. Model upgrades can shift output formatting, tool-call patterns, and edge-case behavior even when overall quality improves.
- Watch your rate limit tier. A new model doesn't reset your usage tier, but traffic shifts during a migration window can behave differently than steady-state — see our rate limits guide if you're seeing more 429s than expected.
- Budget for the September price step-up if you're planning sustained high-volume usage past August 31, 2026.
If you're currently on Opus 4.8 purely for cost reasons rather than because you need the extra reasoning ceiling, Sonnet 5 is worth a side-by-side eval — you may get comparable results at a lower rate.
The Bottom Line
Sonnet 5 is Anthropic's bet that most "we need Opus" workloads were really "we need agentic reliability" workloads — and that a cheaper, more capable Sonnet can absorb them. For new projects, it's the sensible default. For existing Opus workloads, evaluate before you switch. For existing Sonnet 4.6 workloads, there's little reason to wait.