Here is the point upfront: Code-Supernova inside Cline looks like Claude 4.5 Sonnet. The Cline model panel shows a 200k token context window, image input, browser use, prompt caching, and an output price of $15 per million tokens. The copy references “Claude Sonnet 4” and calls out agentic coding and AI agent capabilities. In use, it plans, proposes commands, shows a thinking panel, and it handled my Slay the Spire clone test in the exact style I expect from Claude. One more detail. Although the selector marked it free at one point, my usage incurred charges. The price line item and the feature set are enough to make the call.
What Cline shows for Code-Supernova
Open the Cline model selector and you will see a stealth entry named cline/code-supernova. The model details list:
- Context window: 200,000 tokens
- Supports images
- Supports browser use
- Supports prompt caching
- Input and cache prices surfaced
- Output price: $15.00 per million tokens
The description panel references Claude Sonnet 4 and emphasizes coding, agentic search, and AI agent skills. The UI that appears during runs shows the familiar planning panel with a stepwise plan and command proposals. When I selected Code-Supernova, it produced a clean plan and asked to run commands like mkdir -p to set up the project. That pattern is standard Claude behavior in Cline.
Both Code-Supernova and Claude 4.5 Sonnet show a 200k context window. Exact match.
Pricing tells most of the story
The strongest signal is the $15 per million output token price. That number is a known tell for Claude Sonnet 4.5 output tokens. Many providers have very different output rates. Seeing $15 per million output tokens inside the Code-Supernova details makes the identity very likely. Cline surfaces input and cache prices too, but the output price is the anchor.
One practical note. The selector once labeled it free, but my usage was billed. Assume it is billable unless your dashboard explicitly shows a free tier for that specific model. Model routers and selectors can display outdated badges. Your invoice will not.
Output token price matches at $15 per million. A direct tell.
Why the 200k context window matters for agentic coding
A 200k token window lets you keep much more of the working set live during a task. That means long specifications, multi-file repositories, verbose logs, and plans can sit in context together. Many models claim large windows yet start to drift when you push them into long sequences with browsing and multiple tool calls. Claude Sonnet 4.5 has been solid for plan and act workflows at this scale. Seeing the same window on Code-Supernova is another strong alignment.
For coding, the practical impact is simple. You can move from “toy demos” into tasks that require continuity across several files and steps. You still need to scope and stage work to avoid thrashing, but you can keep more of the plan and the relevant code in memory while the agent works through changes.
Behavior check: plan and act, chain of thought, and coding quality
Code-Supernova shows a planning panel with a clear breakdown of steps. It recommends specific shell commands and proposes cautious file operations. That is the pattern I see from Claude in Cline. In my Slay the Spire clone prompt, it outlined assets, systems, and an implementation path, then proposed the initial directory setup. The overall pacing, the checks before edits, and the way it explains steps all match Claude’s style.
About the visible thinking panel. Cline displays a “thinking” section for agent runs that shows intermediate steps. If you have used Claude in Cline before, the pattern is familiar. Code-Supernova follows this exact style inside the extension.
Feature parity with recent Claude Sonnet updates
- Image support. You can pass screenshots and ask for structured extraction or follow-up edits. That is part of the current Claude Sonnet line.
- Browser use. Plan steps can include search and retrieval. Claude is solid at measured browsing with citations and follow-up filtering.
- Prompt caching. The pricing panel includes cache reads and writes. That aligns with Claude’s token caching to reduce cost on repeated prompts in agent loops.
Stack the features, the behavior, and the price. The profile is the same.
