Claude Code 2.1.284 makes Sonnet 5.5 the default Sonnet model on the Anthropic API, while adding practical changes to permissions, usage visibility and recovery from errors. For people using Claude Code, this is a substantial update with several changes that affect everyday work, not just a longer changelog.
Anthropic Watch analysis
What happened
Anthropic’s release notes list Sonnet 5.5 as the new default Sonnet model on its API, with a one-million-token context window and prices of $2 per million input tokens and $10 per million output tokens. Cache reads are listed at $0.20 per million tokens. The notes also describe changes across Claude Code’s terminal interface, integrations and managed deployments.
Our top picks
- A new default Sonnet model
Sonnet 5.5 is now the default Sonnet model on the Anthropic API, with a one-million-token context window and published token prices. - More deliberate permission prompts
In auto mode, users can allow a read outside the working directories once and still be asked again next time. - Usage costs are easier to see
Claude apps gateway users can see dollar amounts in /usage and the status line, including spending against a monthly limit. - A simpler MCP recovery command
/mcp reconnect all retries every MCP server that failed to connect or needs authentication. - Compaction can try again
If a request is still too long after conversation compaction, Claude Code now compacts again, keeping less of the recent conversation. - Less confusion when the service is struggling
Several fixes address damaged response streams, server errors and unavailable models, including clearer reporting or retries.
Why it matters
The changes give users more visibility over spending and a finer-grained choice when Claude Code requests access beyond its working directories. The new default model and its listed context window and prices also give users concrete details to weigh when deciding how to use the API.
The recovery fixes matter too. A failed connection or an overlong conversation is not a glamorous problem, but it can derail a working session just as effectively as a spectacular one.
Our read
This is a useful, broad update: a model change alongside practical improvements to permissions, usage and reliability. The release notes explain what changed, but do not establish how the new model performs in any particular workflow. Check the API pricing and your own results before treating the new default as an automatic upgrade.
What to watch
- Whether Anthropic publishes more detail on Sonnet 5.5’s capabilities and performance.
- How the revised default affects API costs for existing Claude Code users.
- Whether the new usage display and recovery fixes work smoothly in real deployments.
Discussion spark: Should Claude Code make a new model the default as soon as it ships, or should users have to opt in until they have compared its cost and performance?
Sources and evidence
- v2.1.284 (28 September 2026, 18:02 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.