Cline 3.37: GPT-5.1, Nous Research, Avalon Voice

TL;DR

• GPT-5.1 arrives with optimized prompting and native tool calling for better coding performance
• Nous Research (Hermes 4 70B/405B) joins as a new provider with Cline-specific system prompts
• Avalon speech-to-text now supports Linux; `/deep-planning` gets model-family specific optimization

What Dropped

Cline 3.37 brings three major additions: OpenAI's GPT-5.1 with tailored prompting, Nous Research as a new provider with open-weight Hermes 4 models, and expanded speech-to-text support via Avalon with Linux compatibility. Enterprise configuration options also ship for multi-provider setups and team-wide rule enforcement.

The Dev Angle

GPT-5.1 is live with model-specific prompting and enhanced tool usage. Enable Native Tool Calling in settings for best results—this is the key to unlocking GPT-5.1's coding performance gains. The `/deep-planning` command now uses model-family specific prompts, meaning GPT-5.1, Claude, and Gemini each get optimized instructions for complex planning tasks.

Nous Research's Hermes 4 models (70B and 405B) are now selectable from the provider dropdown. Both come with custom system prompts tuned for Cline's tool patterns. The 70B variant prioritizes speed; the 405B maximizes capability. This gives you a credible open-weight alternative to closed models—useful if you're running inference locally or want vendor independence.

Speech-to-text switched to Aqua Voice's Avalon model for better transcription accuracy. Linux users finally get voice input support (previously macOS and Windows only). Enable it in Speech-to-Text settings.

Enterprise teams get multi-provider configuration, MCP marketplace allowlists, and global rule enforcement across teams—laying groundwork for full enterprise deployment in future releases.

Should You Care?

If you use GPT-4 or GPT-4o: Upgrade to GPT-5.1 immediately. The optimized prompting and native tool calling are material improvements for coding tasks. This is a meaningful step up, not a minor patch.

If you want open-weight models: Hermes 4 is worth testing. The 405B variant competes with frontier closed models; the 70B is fast enough for most workflows. Custom Cline prompting means these models understand your tool usage patterns out of the box.

If you're on Linux: Voice input is now available. Previously a macOS/Windows-only feature, Avalon's transcription accuracy makes this practical for hands-free coding.

If you're running enterprise deployments: The multi-provider and allowlist features are early-stage but signal where Cline is heading. Not production-ready yet, but worth monitoring if you manage teams.

If you use `/deep-planning`: You're now getting model-optimized prompts. No action needed—it just works better.

What's Fixed

Duplicate tool results are gone (fixed existence checking). XML escaping issues in the model content processor are resolved. Commit message generation in the command palette is back. Temperature parameters now convert correctly for OpenAI Compatible providers. HTTP proxy support expanded throughout the codebase. Developers get a new npm run dev script for streamlined terminal development.

Source: Cline