OpenCode Leak Points to Six Unreleased AI Models

OpenCode has quietly tucked references to six AI models into its code, none of which the public can use yet. There was no announcement, no teaser post — just a commit sitting in the repository. The names in it are Kimi K4, GLM 5.5 Flash, GLM 5.4, Muse Spark 1.4, DeepSeek V4.1 Pro, and Qwen 3.8 Max.

The one drawing the most attention is Kimi K4 from Moonshot. Its status in the OpenCode dashboard is still pending and it has zero usage metrics, so nothing is live yet. Word inside the project is that its architecture was built to beat Claude Fable 5.1 and GPT-6 Astra on autonomous coding benchmarks — that part is rumor, not a spec sheet.

What Happened

A commit in the OpenCode repository added references to six models that have not been released to the public. OpenCode made no statement about any of them.

The timing lines up with a busy week elsewhere: OpenAI shipped GPT-6 Luna and Anthropic shipped Claude Fable. If OpenCode's pattern of adding models ahead of launch holds, the Chinese labs behind the other names are getting their reply lineup ready for next week.

Kimi K4 comes from Moonshot, and it is the only one the leak says much about. The rest — GLM 5.5 Flash, GLM 5.4, Muse Spark 1.4, DeepSeek V4.1 Pro, and Qwen 3.8 Max — appear as names without details attached.

Why It Matters

The fight over open-weight models is heating up as late 2026 approaches. DeepSeek V4.1 Pro and Qwen 3.8 Max are both positioned to shake up the developer leaderboards, and a pending Kimi K4 aimed at autonomous coding tasks would land right in the middle of that race.

None of this is confirmed by the labs themselves, so treat the benchmark talk as rumor for now. What is solid is the commit: six model references are in OpenCode's code before any of them exist for the rest of us. Watch the dashboard — if statuses start flipping off pending, the releases are close.