Models

New stealth model: Code-Supernova Gets a Million Token Context Window

The stealth model gets 5x more context. Code-supernova now handles 1 million tokens.

Nick Baumann

26 Sep 2025 — 1 min read

Code-supernova, the free stealth model we launched last week, now supports a 1 million token context window. That's 5x the previous 200k limit.

The expanded context means you can work with entire codebases, extensive documentation, or longer development sessions without context pruning. Everything else remains the same: multimodal support, free access during alpha.

0:00

/0:15

What changes

The model identifier updates to code-supernova-1-million in the model picker. Existing configurations continue working; the model automatically routes to the million token version.

The expanded context helps when:

Working across large monorepos
Debugging with extensive logs
Refactoring interconnected systems
Building with comprehensive API documentation

Access it now

Open Cline settings
Select the Cline provider
Choose code-supernova-1-million
Start coding with 5x more context

The million token window is live now. Share your experience on Reddit or Discord.

0:00

/0:12

happy weekend!

-Nick

Open source progress continues with the release of GLM 4.6 in Cline

The gap between open and closed source AI models is narrowing faster than expected. This week's releases of Sonnet 4.5 and GLM-4.6 exemplify this convergence, with real-world data showing performance differences measured in basis points, not percentage points.

Which local models actually work with Cline? AMD tested them all

After extensive testing, AMD found only specific models work reliably with Cline. Here's your complete guide to choosing the right model for your RAM.

Anthropic's Claude Sonnet 4.5 is Live in Cline

Anthropic's Claude Sonnet 4.5 brings enhanced autonomy and refined communication to Cline, shipping code with less commentary and more focus.

How I stopped repeating myself to Cline

If you're typing the same instructions to your agent every week, you're the process. Write them once as a markdown workflow, invoke with a slash command, and watch the agent execute with verification.