Claude Code v2.1.68: Opus 4.6 defaults to medium effort, reintroduces ultrathink keyword
Claude Code v2.1.68 release notes
Anthropics has released Claude Code v2.1.68 with changes to effort settings and model availability. This update primarily affects how the Opus 4.6 model handles different types of coding tasks.
Key changes in this release
- Opus 4.6 default effort changed: For Max and Team subscribers, Opus 4.6 now defaults to medium effort instead of the previous setting. According to the release notes, "Medium effort works well for most tasks — it's the sweet spot between speed and thoroughness."
- Effort control: Users can change the effort setting anytime using the
/modelcommand. - UltraThink keyword reintroduced: The "ultrathink" keyword has been re-added to enable high effort for the next turn. This allows users to temporarily boost the model's processing for complex tasks without permanently changing settings.
- Model deprecation: Opus 4 and Opus 4.1 have been removed from Claude Code on the first-party API. Users who had these models pinned are automatically moved to Opus 4.6.
The release was published on March 4th by ashwin-ant and includes commit 0b3f7cb. These changes are specific to Claude Code, Anthropics' coding-focused implementation of their Claude models.
📖 Read the full source: GitHub Claude-Code
👀 See Also
Developer Describes Fraud Feeling After First AI-Assisted Pull Request
A developer used Claude Code to create a pull request for Chroma, Hugo's default syntax highlighter, adding ERB syntax highlighting. The PR was approved and merged, but the developer felt like a fraud and experienced worsened impostor syndrome.
Fine-tuned Qwen3 Small Models Outperform Frontier LLMs on Specific Tasks at Lower Cost
Distilled Qwen3 models (0.6B to 8B parameters) matched or beat frontier API models like GPT-5, Gemini, and Claude on 6 out of 9 tasks including function calling and Text2SQL, with cost as low as $3 per million requests versus $378 for comparable performance.
Claude Code v2.1.81 adds bare flag for scripting, fixes authentication and voice mode issues
Claude Code v2.1.81 introduces a --bare flag for scripted -p calls that skips hooks, LSP, and plugin sync, requiring ANTHROPIC_API_KEY or apiKeyHelper via --settings. The release also fixes multiple concurrent session authentication issues, voice mode error handling, and adds --channels permission relay.
OpenClaw LTS Prioritizes Codex Instructions Over AGENTS.md — Breaks Custom Workflows
OpenClaw LTS 2026.6.33 routes all agents through the Codex harness, ignoring AGENTS.md custom instructions. This breaks workflows for users with custom AGENTS.md files.