What happened
On June 30, 2026 Anthropic introduced Claude Sonnet 5 — a new generation of the mid-sized Sonnet that the company calls its most "agentic" yet: the model plans, uses tools (browser, terminal) and works autonomously at a level that until recently required larger and pricier models.
Key facts from the official announcement and the Claude Platform docs:
- API pricing is $3 per million input and $15 per million output tokens, but through August 31, 2026 an introductory price of $2/$10 applies. The API identifier is
claude-sonnet-5. - A 1M-token context window out of the box (both the default and the maximum), with up to 128K output tokens.
- Sonnet 5 became the default model on the Free and Pro plans, is available on Max, Team and Enterprise, and ships in Claude Code — for Free/Pro accounts sessions now start on Sonnet 5.
- Per TechCrunch, on agentic coding Sonnet 5 scores 63.2% against 69.2% for Opus 4.8 and 58.1% for the previous Sonnet 4.6 — a notable jump over its predecessor and close to the flagship.
- It is a drop-in replacement for Sonnet 4.6: just swap the model ID. But mind the details: adaptive thinking is on by default, manual extended thinking is gone (it returns a 400), and non-standard
temperature/top_p/top_kvalues are no longer accepted.
There is an honest catch, too. Sonnet 5 uses a new tokenizer that produces roughly 30% more tokens for the same text. The per-token price has not changed, but the real cost of an equivalent request can go up, and less text fits into that 1M window than before.
Why it matters
The real story is not "another model" but a shift in the price of agentic work. Anthropic positions Sonnet 5 as "performance close to Opus 4.8 but noticeably cheaper." For teams that run a model in a loop (the agent reads code, calls tools, fixes, repeats), the gap between Opus rates and $2/$10 for Sonnet is direct savings on every run.
The 1M context with no asterisks matters separately: a huge window used to be a privilege of top-tier models, and now it is a standard feature at the mid tier. That opens up whole repositories, long logs and big specs within a single session.
But read the news soberly. The ~30% token inflation partly eats the introductory discount (Simon Willison measured roughly 1.4x on English text and ~1.27x on Python code), and the per-token price returns to $3/$15 after August 31. In other words, the cheap window is temporary, and budgets should be recalculated for the new tokenizer rather than carried over from old numbers.
What it means for vibe coding
If you code alongside an assistant, this is your news by default — literally: Claude Code and the Claude apps on Free/Pro now run Sonnet 5, with nothing to switch on.
In practice:
- Fewer reasons to reach for Opus. For everyday "build this feature, fix this bug, run the tests," Sonnet 5 is close to the flagship but cheaper — you can leave the default alone.
- Big context for big tasks. 1M tokens lets you drop many files into a session at once; just remember the new tokenizer — the same amount of code now "weighs" more tokens.
- Check your scripts and integrations. If you have your own API wrapper: drop manual
budget_tokens(adaptive thinking now), remove customtemperature/top_p/top_k(otherwise 400), and recomputemax_tokens— because of the tokenizer, your old limit may truncate the answer. - Watch the pricing calendar. $2/$10 through August 31, then $3/$15. If you run large agentic pipelines, build that jump into your cost estimates.
Bottom line: for a vibe coder this is a free, default upgrade — faster and smarter for the same money — with two caveats: the new tokenizer and the temporary intro price.