Claude-Code v2.1.31 Release: Key Updates and Bug Fixes

This update covers the release v2.1.31 of Claude-Code, which introduces several notable improvements and fixes. Key changes include the addition of session resume hints that guide users on how to continue conversations after exiting. The update also expands support for full-width (zenkaku) spaces input using the Japanese IME in checkbox selections.
Several bugs have been addressed in this release. Notably, issues causing sessions to lock up due to too-large PDF file errors have been fixed, allowing smoother resumption of tasks without needing to start anew. Additionally, bash commands will no longer incorrectly report failures with "Read-only file system" errors when sandbox mode is enabled.
A critical fix was made to the streaming API path where the temperatureOverride parameter was previously ignored. All streaming requests erroneously defaulted to a temperature value of 1, regardless of configuration. This version now respects the specified temperature override, providing more expected behavior.
The update enhances system prompts to better utilize dedicated tools (e.g., Read, Edit, Glob, Grep) instead of relying on bash equivalents. This reduces unnecessary bash command usage. Error messages related to PDF and request sizes have been clarified to show the actual limits of 100 pages and 20MB.
For developers using strict language servers, LSP shutdown/exit compatibility is improved to handle null parameters. Additionally, misleading Anthropic API pricing has been removed from the model selector for users relying on third-party providers like Bedrock, Vertex, and Foundry.
📖 Read the full source: GitHub Claude-Code
👀 See Also

Anthropic Paying SpaceX $15B/Year for Compute Through 2029
SpaceX IPO filing reveals Anthropic pays $1.25B/month through May 2029 for compute. The deal supports AI training on Colossus 1 & 2 facilities.

Adaptive Inference Routing Proposal for AI Query Efficiency
A proposal submitted to Anthropic in April 2026 outlines a five-step system for routing queries to appropriate AI models based on complexity scoring, using simple signals like character count and sentence count before any model inference occurs.

Opus 4.6 excels at research, Gemini 3.1 Pro has better judgment in forecasting benchmark
A benchmark of 1,417 binary forecasting questions separates research and judgment performance: Claude Opus 4.6 leads in agentic research, Gemini 3.1 Pro wins on fixed-evidence calibration. GPT-5.4 and Grok 4.20 show little change between conditions.

Longitudinal study finds AI productivity gains at 10%, not 10x
A longitudinal study tracking 40 companies from November 2024 through February 2026 found AI usage increased by 65% on average, but pull request throughput only increased by 9.97%. The data suggests coding was never the primary bottleneck in software development.