Cline 3.38.0: Gemini 3 Pro & Voice Coding

TL;DR

• Cline 3.38.0 adds Google's Gemini 3 Pro Preview for stronger reasoning and multimodal coding
• New AquaVoice Avalon speech model achieves 97.4% accuracy on coding terms vs. Whisper's 65.1%
• Critical fixes for AWS Bedrock context handling and premature context truncation

What Dropped

Cline 3.38.0 brings two major model additions: Gemini 3 Pro Preview, Google's new flagship for reasoning and code generation, and AquaVoice Avalon, a speech-to-text model trained specifically on engineering language. The release also patches context handling bugs that were causing unnecessary truncation on AWS Bedrock.

The Dev Angle

Gemini 3 Pro shows measurable improvements over Gemini 2.5 Pro on coding benchmarks and agentic workflows. In testing, it maintains coherence across longer tasks, respects the 1M-token context window to keep multi-file edits consistent, and handles frontend generation with design screenshots and documentation in a single conversation. This matters if you're running extended coding sessions where model drift becomes a problem.

For voice-first developers, Avalon is a genuine upgrade. It understands technical terminology that generic speech models butcher—"checkout dev" stays "checkout dev" instead of becoming gibberish. On the AISpeak benchmark of coding terms, Avalon hits 97.4% accuracy compared to Whisper Large v3's 65.1%. If you dictate code or commands regularly, this eliminates the frustration of constant corrections.

Under the hood, the fixes address real pain points: AWS Bedrock users were hitting premature context truncation, and some models (notably Sonnet 4.5) were triggering auto-condensing too early. The team removed the new_task tool from system prompts to prevent "context anxiety" while keeping it available via slash commands.

Should You Care?

If you're already using Cline with Gemini 2.5 Pro, upgrading to 3 Pro is worth testing—the reasoning improvements are real, especially for complex refactoring or multi-file changes. If you use AWS Bedrock, this update fixes a bug that was actively hurting your context window efficiency.

Voice coding with Avalon is a niche win. If you don't dictate code, skip it. If you do, it's a massive quality-of-life improvement. The 97% accuracy on technical terms means fewer interruptions to fix transcription errors.

For most users, this is a solid incremental release. Grab it from your marketplace. If you want deeper context on how Gemini 3 Pro changes your workflow, the team published a full breakdown on Gemini 3 Pro's impact.

Source: Cline