The price hike is cancelled: Sonnet 5 stays at $2/$10
Anthropic cancelled the planned September 1 increase of Claude Sonnet 5 from $2/$10 to $3/$15, so $2/$10 is now the long-term standard price. The official pricing page lists Sonnet 5 at a standard $2 input and $10 output per 1M tokens, with an explicit note that "the planned $3/$15 increase was cancelled" (source).
If you already use Sonnet 5, this means the migration you were planning for late August is no longer needed—you can budget around $2/$10 as the permanent rate.
Pricing comparison before and after the cancellation
| Billing item | Originally planned (from 9/1) | Now (increase cancelled) | Change |
|---|---|---|---|
| Input | $3 / 1M tokens | $2 / 1M tokens | Unchanged |
| Output | $15 / 1M tokens | $10 / 1M tokens | Unchanged |
| Cache write | $3.75 (estimated) | $2.50 | Unchanged |
| Cache read | $0.30 (estimated) | $0.20 | Unchanged |
The pricing comparison conclusion is simple: the hike was called off, and Sonnet 5's input, output and cache prices all stay at the introductory level.
Four-model pricing comparison: Sonnet 5 sits at the sweet spot
Place Sonnet 5 next to the other three Claude mainline models, and it lands exactly at the cheapest tier with 1M context and adaptive thinking. The official lineup runs Haiku 4.5, Sonnet 5, Opus 5, Fable 5 from low to high (source):
| Model | Input | Output | Cache write | Cache read | Context | Max output |
|---|---|---|---|---|---|---|
| Claude Haiku 4.5 | $1 | $5 | $1.25 | $0.10 | 200K | 64K |
| Claude Sonnet 5 | $2 | $10 | $2.50 | $0.20 | 1M | 128K |
| Claude Opus 5 | $5 | $25 | $6.25 | $0.50 | 1M | 128K |
| Claude Fable 5 | $10 | $50 | $12.50 | $1 | 1M | 128K |
This pricing comparison shows Sonnet 5 costs one-quarter of Opus 5's input and one-fifth of Fable 5's, while keeping the same 1M context and 128K output. For daily production, team backends and coding agents, Sonnet 5 is the best value tier.
Pricing comparison against rivals: worth switching?
Benchmarking Claude Sonnet 5 against mainstream APIs, it is pricier than domestic flagships, but the cancelled hike restores its value in the "1M context + top-tier coding" combination. Using input/output list prices:
| API | Entry/workhorse (input/output) | Flagship (input/output) |
|---|---|---|
| Claude API | Sonnet 5 $2/$10 | Fable 5 $10/$50 |
| DeepSeek | Flash ~$0.08/$0.18 | Pro ~$0.435/$0.87 |
| OpenAI API | Luna ~$0.10/$0.60 | Sol ~$5/$30 |
This pricing comparison shows Sonnet 5 is more expensive than DeepSeek and OpenAI entry tiers, but its selling point is 1M context plus Anthropic's coding and agent quality. For Chinese-first lightweight tasks, DeepSeek is cheaper; when you need Claude-grade long-context coding and adaptive thinking, Sonnet 5 is the cheapest entry. Rival figures are list prices—confirm on each provider's pricing page.
Official entry point, API key and integration config
The cancellation changes nothing about how you access the API—Claude API's entry point and key flow are unchanged. Key points:
- Create the API key in the Anthropic Console (console.anthropic.com), separate from Claude Pro/Max subscriptions and billed per token.
- The model id stays
claude-sonnet-5, so existing integrations need no edits—only the unit price remains $2/$10. - Savings come from two layers: prompt cache reads at 0.1x the input price ($0.20/M) and Batch API at about 50% off, completing asynchronously within 24 hours.
- For a fixed monthly budget use the Claude subscription; for pay-as-you-go, top up and use the Console API.
Things to watch
-
This $2/$10 is the API's Sonnet 5, not a subscription tier. Pro/Max subscriptions are a separate billing system and do not share this rate—do not mix API token prices with subscription monthly fees.
-
Caching and Batch are the real savings levers. Sonnet 5's unit price is fixed; actual cost depends on cache-hit rate for repeated context and whether async jobs can go through Batch at half price.
-
Legacy models still appear in the table. Sonnet 4.6, Opus 4.7 and others may show up; new integrations should use Sonnet 5 directly.
-
Regional and cloud endpoints carry premiums. US-only inference bills at 1.1x, and Bedrock/Vertex regional endpoints may add 10%—model cost by your actual deployment region.