Anthropic Extends Claude Code Usage Boost to August 19 — And Sonnet 5 API Pricing Jumps 50% on September 1
Claude Code Pro/Max limits stay 50% higher through August 19, 2026. Sonnet 5 API intro pricing ends August 31 — and a tokenizer change makes the real cost rise bigger than it looks.
Two separate Anthropic policy updates are running on overlapping timelines right now, and the second one carries a cost sting that’s easy to underestimate. Here’s what’s happening and what you should do before the deadlines hit.
Claude Code’s 50% Usage Boost Has Been Extended (Again)
If you’re on a Pro, Max, Team, or seat-based Enterprise plan, your Claude Code weekly limits are currently running 50% higher than the baseline. Anthropic confirmed on July 18 via the @ClaudeDevs account that this temporary boost runs through August 19, 2026.
This is worth putting in context. Back in May 2026, Anthropic permanently doubled Claude Code’s 5-hour limits for those same plan tiers. The current 50% weekly boost sits on top of that permanent change. So the floor is already higher than it used to be, and for now you’re getting a further uplift on top of that.
The boost has been extended more than once, so it may well continue past August 19. But the official end date is what it is, and you should plan your sprint work accordingly. If you have heavy Claude Code sessions scheduled, front-loading them before mid-August is the sensible move.
One housekeeping note: some users have reported the increased limits not showing up correctly in /usage. If your counter looks lower than expected, the underlying limits are still in effect.
Sonnet 5 API Pricing: The September 1 Cliff Is Bigger Than It Looks
This is the update that deserves more attention.
Claude Sonnet 5 launched on June 30, 2026 with introductory API pricing of $2 per million input tokens and $10 per million output tokens. That promotional rate expires August 31, 2026. From September 1, the standard rates apply: $3 input / $15 output per million tokens.
That’s a clean 50% increase on both sides. Straightforward enough. But there’s a second factor that makes the effective cost rise larger than the list-price change suggests.
Sonnet 5 ships with an updated tokenizer, inherited from the Opus 4.7 generation, that maps the same content to 1.0x to 1.35x more tokens compared to Sonnet 4.6. For casual conversational prompts, the difference is small. For dense technical content, large codebase inputs, long agent transcripts, tool-call histories, or browser observation logs, you can be closer to the 1.35x end.
What this means in practice: on September 1, you’re not just paying 50% more per token. You’re also potentially feeding more tokens through for the same prompts. The two effects compound. An identical workload that ran fine at Sonnet 4.6 rates could cost meaningfully more than a simple 50% uplift calculation would predict.
Anthropic framed the introductory pricing as making the Sonnet 4.6 to Sonnet 5 transition roughly cost-neutral. At the promo rate, that’s broadly true. At standard rate, the tokenizer change tips the balance.
What You Should Do Before August 31
If you’re an API developer using Sonnet 5:
Now, while costs are still at the promotional rate, is the right time to benchmark your actual workloads. Pull your most common prompt templates, run them through the token counter in the Anthropic Console, and record actual Sonnet 5 token counts. Do not use Sonnet 4.6 counts as your baseline for prose-heavy or code-heavy content. They will undercount by up to 35%.
Once you have real numbers, model out your September spend at $3/$15. The goal is no surprises on your first post-August invoice.
Also check your prompt caching setup. Because cached prefix sizes can differ between Sonnet 4.6 and Sonnet 5 due to the tokenizer change, caching behaviour that worked well before may need reconfiguration.
A quick note on breaking API changes: if you’re migrating code to Sonnet 5, there are a few hard failures to handle. Setting temperature, top_p, or top_k to non-default values now returns a 400 error. The manual extended thinking parameter (thinking: {type: "enabled", budget_tokens: N}) also returns a 400; you need to use the new effort parameter instead.
If you’re a Claude Code Pro or Max subscriber:
The 50% weekly boost is automatic. No action needed to keep it through August 19. Just be aware that if Anthropic doesn’t extend it again after that date, your weekly limits will step down to the permanent (still-doubled) baseline. That baseline is roomier than the old pre-May limits, so it won’t feel like a cliff, but it’s worth knowing.
The Pricing Picture at a Glance
| Period | Input | Output |
|---|---|---|
| Now through August 31, 2026 | $2 / M tokens | $10 / M tokens |
| September 1, 2026 onwards | $3 / M tokens | $15 / M tokens |
The practical takeaway: the next six weeks are your cheapest window to run Sonnet 5 workloads. Use that time to understand your real token consumption under the new tokenizer before the standard rate kicks in.