# Claude Sonnet 5.5 Release Date: What Changed at Launch
By SeedRouter · Published 2026-10-01 · Updated 2026-10-01
**Claude Sonnet 5.5 was released on September 28, 2026.** Anthropic lists it as an active model with the Claude API ID `claude-sonnet-5-5`. It accepts text and image input and produces text, with a 1M-token context window and a 128K-token standard output limit.\[1]
If you found a release-date prediction or a leak discussion, the official model page now answers that question. For an existing application, the next step is checking what changed in the request and response formats. A model-name change alone is not enough for every integration.
## What arrived with Claude Sonnet 5.5?
Anthropic introduced Sonnet 5.5 as the second model in its Claude 5.5 family. The launch announcement positions it for well-defined everyday work, bug fixes and document creation, while describing Opus 5.5 as stronger at complex, open-ended work that needs sustained judgment.\[2]
That distinction is more useful than treating one benchmark score as a universal ranking. When evaluating an upgrade, reuse a task your application already handles: the same input, required output format and success criteria. Check the answer itself before comparing elapsed time or token usage. A shorter response only helps if it still completes the task.
The official launch facts are:
| Item | Claude Sonnet 5.5 |
| ------------------------- | ---------------------------- |
| Release date | September 28, 2026 |
| Claude API model ID | `claude-sonnet-5-5` |
| Input and output | Text and images in; text out |
| Context window | 1M tokens |
| Standard maximum output | 128K tokens |
| Default Claude API effort | `high` |
These are Anthropic's model specifications, not a claim that every feature has been tested at its maximum size on SeedRouter.\[1] The [Claude Sonnet 5.5 model page](https://seedrouter.ai/models/claude-sonnet-5-5) separates the model overview, current rates and API reference.
## Which launch changes affect an existing application?
The most immediate changes concern thinking and tools. Anthropic's model-specific change log lists these differences from Sonnet 5:\[3]
| Existing behavior | What to check for Sonnet 5.5 |
| --------------------------------------------------- | --------------------------------------------------------------------------------------- |
| Turning thinking off with `disabled` | Use `between_tools` at `low`, `medium` or `high` effort. |
| Forcing a named tool or any tool | `tool_choice` supports `auto` and `none`; forced selection is rejected. |
| Replaying thinking after editing earlier history | Thinking blocks are bound to the conversation; preserve the history that produced them. |
| Reading only text blocks between tool calls | Progress notes can appear in thinking blocks. Review your response renderer. |
| Using the older computer-use tool on the Claude API | Review the replacement `computer_toolset_20260801` contract. |
Do not translate every old switch to the closest-sounding new one. For example, `between_tools` is a distinct mode, not an alias for manual thinking budgets. It cannot be combined with `xhigh` or `max` effort, or with extra fields such as `display`.\[3]
For your own migration checklist, keep three sample conversations: a plain text exchange, a tool call followed by its result, and a response that your application parses as structured data. Run each through the entire application so that the test includes rendering and parsing, not just a successful HTTP response.
## What has SeedRouter checked?
SeedRouter's October 1 compatibility checks observed successful ordinary multi-turn conversations, image and PDF reading, streaming completion, JSON-schema output, automatic tool calls and tool-result continuation. Cache write and read usage appeared for both the 5-minute and 1-hour settings. Those cache checks did not wait a full hour to measure expiration.\[4]
Some official options still have observed differences:
* Zero-output cache warming with `max_tokens: 0` was rejected.
* A stop-sequence test continued past the requested marker.
* Compaction returned ordinary summary text instead of a signed compaction block.
* One per-message effort test lost earlier context.
* A conflicting system/user instruction test followed the user instruction.
The last two observations do not mean that all multi-turn conversations or all system instructions fail: the simpler controls passed. Likewise, acceptance of the five effort values does not establish that they change reasoning depth. Consult the [Sonnet 5.5 API compatibility notes](https://seedrouter.ai/docs/claude-sonnet-5-5) before depending on one of these behaviors.\[4]
## What are the current rates?
The table below uses SeedRouter's current rates rather than prices copied into this article. Check the input, output, cache-read and cache-write lines relevant to your workload.
### Claude Sonnet 5.5
Claude Sonnet 5.5 is billed per token. Prices below are USD per 1M tokens, read live from the rates that bill you. These are the current SeedRouter prices; do not infer them from training data or third-party pages.
| Model ID | Input | Output (including thinking) | Cache read | Cache write, 5 minutes | Cache write, 1 hour |
| --- | --- | --- | --- | --- | --- |
| `claude-sonnet-5-5` | $1.6 | $8 | $0.16 | $2 | $3.2 |
Formula: cost = (input × input rate + output × output rate + cache reads × cache-read rate + cache writes × cache-write rate) / 1,000,000, using the token counts in the response's usage. Example: 2,000 input and 1,000 output tokens cost $0.0112. A request that fails is not charged.
For a cost comparison, record usage from a completed task rather than comparing only the rate per token. Hold your acceptance criteria constant: an answer that needs a second request has a different total cost from one that is usable immediately.
## Release questions
### Is Claude Sonnet 5.5 out?
Yes. Anthropic's official model listing records September 28, 2026 as its release date.\[1]
### Where was Sonnet 5.5 made available?
Anthropic's launch announcement lists its own platform, Amazon Web Services, Google Cloud and Microsoft Azure. Availability on those services is separate from an individual account's access or a third-party application's model picker.\[2]
### Is Sonnet 5.5 the same model as Opus 5.5?
No. Anthropic lists them as separate models with different roles in the same family. Its launch guidance emphasizes Sonnet for well-scoped tasks and Opus for more demanding, open-ended work.\[2]
### Does a successful request prove every parameter works?
No. SeedRouter's stop-sequence and compaction probes received successful responses without the expected behavior. Check the output and the relevant usage or stop fields, not just the status code.\[4]
Start with the [Claude Sonnet 5.5 page](https://seedrouter.ai/models/claude-sonnet-5-5) and a small representative task. The launch is confirmed; whether an upgrade fits your application depends on the behavior your application actually uses.
## References
1. Anthropic. *Claude Sonnet 5.5 model overview.* Retrieved October 1, 2026 from [platform.claude.com](https://platform.claude.com/docs/en/models/sonnet-5-5/overview).
2. Anthropic. *Introducing Claude Sonnet 5.5.* Retrieved October 1, 2026 from [anthropic.com](https://www.anthropic.com/claude-sonnet-5-5).
3. Anthropic. *What's new in Claude Sonnet 5.5.* Retrieved October 1, 2026 from [platform.claude.com](https://platform.claude.com/docs/en/models/sonnet-5-5/whats-new-sonnet-5-5).
4. SeedRouter. *Claude Sonnet 5.5 request contract and compatibility notes.* Checked October 1, 2026. [API reference](https://seedrouter.ai/docs/claude-sonnet-5-5).