Claude Desk / Sonnet 5 price lock / Source-backed briefing / 2026-08-14
← Back to The AGI Times
The AGI Times
Source Notes Desk
Small-business office desk with invoices, a calculator, and a laptop. No logos.
Claude Desk / Canada / 2026-08-14

Anthropic Locked Claude Sonnet 5 At $2 And $10

The September hike is off. That is the good budget news. The other line still matters: the same page of text can count as more tokens than it did on older Sonnet models.

On August 10, 2026, Anthropic made Claude Sonnet 5's launch price the standard price. It stays $2 per million input tokens and $10 per million output tokens. The $3 and $15 rates that were set for September 1 will not take effect.

Claude Sonnet 5 pricingAnthropic API priceClaude Code costAlberta AI automationprompt caching
Fast source checkSource check: Anthropic's June 30 launch post, updated August 10. The Claude Platform pricing page and the August 10 release notes say the same thing. The hike is cancelled, not delayed.

What Anthropic actually changed

Sonnet 5 shipped on June 30 as the default model on Free and Pro, and as an option on Max, Team, and Enterprise. The API name is claude-sonnet-5. The $2 / $10 rate was sold as introductory pricing through August 31. That sale is now the list price.

The official price table still lists Claude Sonnet 4.6 at $3 / $15. Batch processing on Sonnet 5 is $1 / $5. Prompt cache writes are $2.50 for five minutes and $4 for one hour. Cache hits are $0.20. Those are Anthropic's published USD rates, not a Canadian shop's invoice.

The sticker cut is not the whole bill

Anthropic says Sonnet 5 uses a newer tokenizer. The June 30 release notes put the increase at about 30 percent more tokens for the same text. The launch post footnote gives a wider range: roughly 1.0 to 1.35 times, depending on the content. The exact count depends on the file.

That means a team that compared $2 / $10 with the old $3 / $15 on Sonnet 4.6 cannot treat the job as one-third cheaper. Measure tokens on a real quote draft, a tender pack, or a support thread. Then price the finished task, not the brochure rate.

Two other line items sit on the same pricing page. US-only inference on Claude 4.6 and later carries a 1.1 times multiplier. Priority Tier is not available on Sonnet 5. If a contractor promised you that add-on, ask again.

What a local operator should do

If you already budgeted a September Claude hike, you can keep the $2 / $10 plan and spend the slack on review time. If you run Claude Code or Cowork on client files, keep those jobs off a public chat window. A cheaper model does not change who owns the quote.

Do not rip out Gemini, Grok, or Codex because Anthropic held a price. Pick one internal task, keep a human check, and compare error rate and token count for two weeks. Record the prompt, the output, and the human fix.

Opcelerate recommendationRebuild the cost sheet around finished work: tokens in, tokens out, cache hits, and minutes of human review. Use the locked $2 / $10 rate. Do not assume the September 1 hike is coming back unless Anthropic publishes a new change.