Qwen 3.6/3.5 Plus arrive in Go

OpenCode’s Go subscription is adding Qwen 3.6 Plus and Qwen 3.5 Plus, expanding the paid model lineup for developers who use the coding agent through its hosted provider. (x.com) The rollout centers on two things developers ask for in coding tools: image input and tighter privacy controls. OpenCode said the new Qwen models in Go support images, and the company’s docs describe Go as a beta subscription tier that works through an API key inside the terminal interface. (x.com) (opencode.ai) OpenCode Go costs $5 for the first month and $10 a month after that. The docs say the service is aimed at international users and runs models from infrastructure in the United States, Europe, and Singapore. (opencode.ai 1) (opencode.ai 2) Go is OpenCode’s low-cost hosted lane for open models, while Zen is its pay-as-you-go gateway and Enterprise is the self-controlled setup for companies that want traffic routed through their own approved systems. OpenCode’s Enterprise docs say the company does not store code or context data, except when users explicitly turn on the optional share feature. (opencode.ai 1) (opencode.ai 2) (opencode.ai 3) That context matters because OpenCode has been pushing subscriptions when free access runs out. In an April 2026 GitHub issue, users reported seeing a message that Qwen 3.6 Plus Free had ended and that continued access required subscribing to OpenCode Go. (github.com) The image support also lands after recent friction around Qwen inside OpenCode. A GitHub issue filed four days ago said Qwen 3.6 Plus in OpenCode 1.3.13 threw an error when users attached an image, while Qwen 3.5 Plus handled the same input normally. (github.com) Qwen is Alibaba’s model family, and the company’s February 15, 2026 post on Qwen 3.5 described it as a native vision-language model, meaning one model can process text and pictures together instead of bolting image handling on later. Alibaba said the open-weight Qwen3.5-397B-A17B activates 17 billion parameters out of 397 billion total during each pass to keep inference more efficient. (qwen.ai) A separate April 2, 2026 listing for Qwen 3.6 Plus on OpenRouter describes the newer model as having a 1 million-token context window and stronger coding and multimodal performance than the 3.5 series. OpenRouter lists weighted average pricing at about $0.50 per million input tokens and $3.00 per million output tokens across providers, which helps explain why OpenCode is also pitching Qwen 3.5 Plus as the cheaper option. (openrouter.ai) OpenCode’s own Go docs still show a model list led by GLM, Kimi, MiMo, and MiniMax models, and say that list can change as the team tests additions. This Qwen update puts two Alibaba models into the paid Go lane just as OpenCode’s April 9 changelog added prompts nudging free users toward a Go subscription when they hit limits. (opencode.ai) (opencode.ai) The immediate test is whether Qwen in Go gives developers a steadier path than the recent free-tier and image-input glitches. For OpenCode, the pitch is now simple: pay $10 a month, get hosted open models that can read code and pictures, and keep more control over where the data goes. (opencode.ai) (github.com)