Together adds hosted Muse Spark Open access, with separate input, output and cache prices

At a glance

• The source advertises a 1M-token context window for long coding tasks, multi-file changes and project context. It also describes planning, tool use, subagents, computer use and intent commentary. These are provider descriptions, not our benchmark results or a guarantee that every repository fits into one request.
• The model is described as accepting text, images, audio and audio-video. The post supplies a serverless SDK example and mentions dedicated infrastructure, but we did not run either. Model capabilities described in the announcement do not independently verify every modality or tool in each deployment.
• The published usage prices are US$1.25 per million input tokens, US$4.25 per million output tokens and US$0.15 per million cached input tokens. They are distinct billing categories, not a monthly plan or a promise that all input receives the cached rate. A playground link does not establish free usage.
• Factory and Ollama are named as early-access customers offering the model to users from day zero. Their statements praise tool calling and long context, but they are customer testimonials. They do not establish a general model ranking, universal reliability or that the full model runs locally on any consumer device.