Claude Sonnet 4.6 launches as the new flagship balanced model, replacing Sonnet 4.5 as the default across Claude Code and the API. It supports adaptive thinking, 1M token context window (beta), and free code execution when paired with web tools.
Code execution, web fetch, web search, memory tool, programmatic tool calling, and tool search tool all graduate to general availability — no beta headers required.
Web search and web fetch now support dynamic filtering (beta): Claude can write and execute code to filter results before they reach the context window, improving accuracy and reducing token costs.
Claude Code 2.1.45 adds spinnerTipsOverride customization, SDKRateLimitInfo types, and fixes Agent Teams failures on Bedrock/Vertex/Foundry.
The legal and compliance docs now include an explicit Usage Policy section clarifying that OAuth tokens from Free/Pro/Max plans cannot be used in third-party products or the Agent SDK.
Added support for Claude Sonnet 4.6 as the new default model
Added support for reading enabledPlugins and extraKnownMarketplaces from --add-dir directories
Added spinnerTipsOverride setting to customize spinner tips — configure tips with an array of custom tip strings, and optionally set excludeDefault: true to show only your custom tips instead of the built-in ones
Added SDKRateLimitInfo and SDKRateLimitEvent types to the SDK, enabling consumers to receive rate limit status updates including utilization, reset times, and overage information
Existing feature improvements
Improved startup performance by removing eager loading of session history for stats caching
Improved memory usage for shell commands that produce large output — RSS no longer grows unboundedly with command output size
Improved collapsed read/search groups to show the current file or search pattern being processed beneath the summary line while active
[VSCode] Improved permission destination choice (project/user/session) to persist across sessions
Major bug fixes
Fixed Agent Teams teammates failing on Bedrock, Vertex, and Foundry by propagating API provider environment variables to tmux-spawned processes
Fixed sandbox "operation not permitted" errors when writing temporary files on macOS by using the correct per-user temp directory
Fixed Task tool (backgrounded agents) crashing with a ReferenceError on completion
Fixed autocomplete suggestions not being accepted on Enter when images are pasted in the input
Fixed skills invoked by subagents incorrectly appearing in main session context after compaction
Fixed excessive .claude.json.backup files accumulating on every startup
Fixed plugin-provided commands, agents, and hooks not being available immediately after installation without requiring a restart
Updated default primary model to claude-sonnet-4-6 (was claude-sonnet-4-5-20250929), reflected in the models table and example environment variable exports. [line 165] [Source]
Added a new Usage Policy section covering acceptable use and authentication/credential restrictions. OAuth tokens from Free, Pro, or Max accounts are exclusively for Claude Code and Claude.ai and cannot be used in third-party tools or the Agent SDK. Developers building on Claude must use API key authentication. [lines 20-35] [Source]
Updated sonnet alias description to reflect that it now points to Sonnet 4.6. [line 20] [Source]
Clarified default model behavior: Max and Team Premium now default to Opus 4.6; Pro and Team Standard now default to Sonnet 4.6. [lines 131-132] [Source]
Updated all model identifier examples from claude-sonnet-4-5-20250929 to claude-sonnet-4-6 in telemetry attributes and event log schema. [lines 313, 322, 389, 407]
Adaptive thinking is now supported on Claude Sonnet 4.6 in addition to Opus 4.6. [line 4] [Source]
Clarified that extended thinking with budget_tokens is deprecated on both Opus 4.6 and Sonnet 4.6 but remains functional. [line 16] [Source]
Clarified interleaved thinking availability by mode: automatic with adaptive thinking on both models; supported via beta header with manual mode on Sonnet 4.6; not available with manual mode on Opus 4.6. [lines 124-131] [Source]
Updated the model recommendation table: Sonnet 4.6 replaces Sonnet 4.5 with the description "Frontier intelligence at scale—built for coding, agents, and enterprise workflows." [line 58] [Source]
Code execution is now generally available — beta header (code-execution-2025-08-25) is no longer required. All example requests updated accordingly. [line 2] [Source]
Code execution is now free when used with web search or web fetch (web_search_20260209 or web_fetch_20260209). [line 8] [Source]
Added Claude Sonnet 4.6 to the supported models table. [line 21] [Source]
Added platform availability section: available on Claude API and Microsoft Azure AI Foundry; not currently available on Amazon Bedrock or Google Vertex AI. [lines 35-42] [Source]
Clarified that code execution is not ZDR-eligible; data is retained per standard policy. [line 14] [Source]
Added code execution usage tracking example showing server_tool_use.code_execution_requests in the response. [lines 568-578] [Source]
Claude Sonnet 4.6 added to the 1M token context window support list (beta for tier 4+ orgs). [line 90] [Source]
Sonnet 4.6 now features context awareness (tracks remaining token budget during long conversations), joining Sonnet 4.5 and Haiku 4.5. [line 122] [Source]
The effort parameter is now supported on Claude Sonnet 4.6 in addition to Opus 4.6 and Opus 4.5. [line 7] [Source]
Added recommended effort levels for Sonnet 4.6: medium (recommended default), low for latency-sensitive workloads, high for maximum intelligence. [lines 39-46] [Source]
Added Claude Sonnet 4.6 to the supported models list — supports both manual extended thinking with interleaved mode and adaptive thinking. [line 15] [Source]
Updated interleaved thinking documentation: Opus 4.6 automatically enables it with adaptive thinking (beta header deprecated); Sonnet 4.6 supports it via beta header with manual mode or automatically with adaptive thinking. [lines 277-285] [Source]
Updated comparison table to include Sonnet 4.6 column. [line 505] [Source]
Added a comprehensive new section: Migrating to Claude Sonnet 4.6, covering breaking changes (assistant prefilling removed, JSON escaping differences, tool version updates), recommended changes (remove fine-grained tool streaming beta header, migrate output_format), and detailed migration paths for extended thinking, adaptive thinking, and effort levels. [lines 158-330] [Source]
Clarified that the interleaved-thinking-2025-05-14 beta header deprecation applies to Opus 4.6 only; Sonnet 4.6 continues to support it. [line 43] [Source]
Claude Sonnet 4.6 replaces Sonnet 4.5 in the latest models comparison table. API ID is claude-sonnet-4-6. Adaptive thinking is now supported. Reliable knowledge cutoff is Aug 2025; training data cutoff is Jan 2026. [lines 15-29] [Source]
Added Claude Sonnet 4.6 pricing: $3/$15 per MTok (input/output), same as Sonnet 4.5. [line 19] [Source]
Added Sonnet 4.6 to Batch API pricing table at $1.50/$7.50 per MTok. [line 98] [Source]
Added Sonnet 4.6 to long context pricing (supported with 1M token context window beta). [line 111] [Source]
Added Sonnet 4.6 to tool use system prompt token counts table. [line 179] [Source]
Expanded code execution pricing section to clarify free usage with web tools and added usage tracking example showing server_tool_use.code_execution_requests. [lines 210-230] [Source]
Programmatic tool calling is now generally available — beta header (advanced-tool-use-2025-11-20) is no longer required. All example requests updated accordingly. [line 2] [Source]
Added Claude Sonnet 4.6 to the supported models table. [line 18] [Source]
Clarified programmatic tool calling is not ZDR-eligible. [line 8] [Source]
Added Claude Sonnet 4.6 to the supported models list and pricing table. [lines 92, 121]
Added a note that prompt caching may be suitable for ZDR customers (it stores KV cache representations and hashes, not raw prompt text). [line 7] [Source]
Added February 17, 2026 release notes: Sonnet 4.6 launch, code execution now free with web tools, web search and programmatic tool calling going GA, and additional tools graduating to GA. [lines 9-15] [Source]
Web fetch tool is now generally available — beta header (web-fetch-2025-09-10) is no longer required. All example requests updated. [line 7] [Source]
Added Claude Sonnet 4.6 to the supported models list. [line 29] [Source]
Added new web_fetch_20260209 tool version supporting dynamic filtering (beta): use with code-execution-web-tools-2026-02-09 beta header to let Claude filter fetched content using code execution before it reaches the context window. [lines 49-92] [Source]
Clarified web fetch is ZDR-eligible. [line 10] [Source]
Web search tool is now generally available (no beta header required). [line 4] [Source]
Added Claude Sonnet 4.6 to the supported models list. [line 19] [Source]
Added new web_search_20260209 tool version supporting dynamic filtering (beta): Claude can write and execute code to post-process search results before they reach context, improving accuracy and reducing token costs. Requires code-execution-web-tools-2026-02-09 beta header. [lines 34-75] [Source]
Clarified web search is ZDR-eligible. [line 8] [Source]
Added Claude Sonnet 4.6 (claude-sonnet-4-6) to the models table — supports 200K context (with 1M token beta), 64K max output, extended thinking, and adaptive thinking. [line 12] [Source]
Documented adaptive thinking support for Sonnet 4.6 including effort parameter recommendations. [lines 31-48] [Source]
Added new section: Code execution is now free with web tools when web_search_20260209 or web_fetch_20260209 is included in requests. [lines 42-45] [Source]
Added new section: Improved web search and web fetch with dynamic filtering (beta) — available for Opus 4.6 and Sonnet 4.6 via code-execution-web-tools-2026-02-09 beta header. [lines 48-52] [Source]
Added new section: Tools graduating to general availability: code execution, web fetch, programmatic tool calling, tool search, tool use examples, and memory tool. [lines 54-63] [Source]
Clarified that interleaved-thinking-2025-05-14 beta header is deprecated on Opus 4.6 but continues to be supported on Sonnet 4.6. [lines 101-103] [Source]