What happens in Claude Code 2.1.286 when the API refuses my model?

Before Claude Code 2.1.286, every turn failed when the Anthropic API refused the model that your default setting or a model alias resolved to. In 2.1.286 Claude Code retries once on the previous model of the same tier, and the fallback notice says when that drop also cut the context window from 1M to 200K tokens.

Did Claude Code 2.1.286 fix lost sessions after a crash?

Yes. Claude Code 2.1.286 fixes claude --resume and --continue sometimes losing every turn after a batch of parallel tool calls when the earlier session had crashed or been killed. It also fixes cloud sessions with very large histories that never woke up because the container stopped while the transcript was still loading.

What changed in Claude Code's --bare mode?

In Claude Code 2.1.286, --bare connects only the MCP servers named on the command line, sends the model no system reminders, and starts no background tasks. Under --bare, a shell command that reaches its timeout now stops instead of moving to the background, which makes scripted runs more predictable.

Was the Claude apps gateway spend meter wrong before 2.1.286?

Yes, in two ways. Before Claude Code 2.1.286 the gateway's spend meter priced 1-hour prompt cache writes at the cheaper 5-minute rate, and on streamed turns that ran a server-side tool such as web search it counted only the first model call's input tokens. Both are fixed.

Anthropic · 2026-09-30 · major

Claude Code 2.1.286 — counted permission prompts and model-refusal retries

Claude Code 2.1.286 numbers stacked permission prompts ("2 of 5"), retries on the previous model when the API refuses the default, and fixes resumed sessions that lost turns, secret leaks in logs and gateway spend pricing.

Claude Code numbers piled-up permission prompts, survives model refusals, and stops leaking secrets in logs.

Quick facts

Version	v2.1.286
Maker	Anthropic
Released	2026-09-30
Retry limit	At most 14 requests per failing model call (default settings)
Changed flag	--bare now connects only the MCP servers named on the command line
New in VS Code	Bookmarks side panel

What is it?

Claude Code 2.1.286 shows a count such as "2 of 5" when several permission requests stack up. When the Anthropic API refuses the model your default or alias points to, every turn used to fail; now the CLI retries once on the previous model of the same tier. The VS Code extension adds bookmarks for saving Claude's answers in a side panel.

How does it work?

Retries are now counted per model call rather than per request, and --bare becomes a truly minimal mode with no system reminders and no background tasks. A large batch of redaction fixes masks percent-encoded Bearer tokens, secrets with zero-width characters in their names, and URL passwords that contain punctuation. Plugin installs now refuse npm sources that are git repos or folders.

Why does it matter?

The resume fix matters most for long agent runs: claude --resume and --continue could lose every turn after a batch of parallel tool calls if the session had crashed. Teams on the Claude apps gateway also get correct spend numbers, because 1-hour prompt-cache writes were priced at the cheaper 5-minute rate.

Who is it for?

Claude Code users and team admins

Frequently asked questions

What happens in Claude Code 2.1.286 when the API refuses my model?: Before Claude Code 2.1.286, every turn failed when the Anthropic API refused the model that your default setting or a model alias resolved to. In 2.1.286 Claude Code retries once on the previous model of the same tier, and the fallback notice says when that drop also cut the context window from 1M to 200K tokens.
Did Claude Code 2.1.286 fix lost sessions after a crash?: Yes. Claude Code 2.1.286 fixes claude --resume and --continue sometimes losing every turn after a batch of parallel tool calls when the earlier session had crashed or been killed. It also fixes cloud sessions with very large histories that never woke up because the container stopped while the transcript was still loading.
What changed in Claude Code's --bare mode?: In Claude Code 2.1.286, --bare connects only the MCP servers named on the command line, sends the model no system reminders, and starts no background tasks. Under --bare, a shell command that reaches its timeout now stops instead of moving to the background, which makes scripted runs more predictable.
Was the Claude apps gateway spend meter wrong before 2.1.286?: Yes, in two ways. Before Claude Code 2.1.286 the gateway's spend meter priced 1-hour prompt cache writes at the cheaper 5-minute rate, and on streamed turns that ran a server-side tool such as web search it counted only the first model call's input tokens. Both are fixed.

Try it

npm i -g @anthropic-ai/claude-code@2.1.286