AI

Mistral launches Responses API preview and adds two low-cost models

Wednesday, October 28, 2026Read Original

Details

  • Mistral AI launched the Responses API in public preview for building agent applications on its models.
  • Each response separates the model’s answer, reasoning, and tool calls into distinct items, simplifying application-level orchestration.
  • The API supports text, image and file inputs, structured outputs, streaming, and client-executed function calls.
  • DeepSeek V4.1 Flash entered public preview on Mistral AI Cloud Inference for long-context coding and agentic workloads.
  • GLM 5.3 Flash became generally available for subagents, tool calls, coding, and other latency-sensitive tasks.
  • Both models are served in the European Union and United States through Mistral AI Cloud Inference.
  • The API uses an OpenAI-compatible Responses format, while conversation history remains stateless and must be managed by the application.

Impact

The release expands Mistral’s appeal to developers building multi-step agents by pairing an OpenAI-compatible interface with hosted third-party models. It could lower inference costs and simplify migrations, particularly for coding and tool-use workloads, while putting pressure on rival platforms that compete on agent APIs, model choice, and regional availability. Hosting DeepSeek and GLM models also strengthens Mistral’s role as an inference marketplace rather than a single-model provider.

Rift Dispatch
Mistral launches Responses API preview and adds two low-cost models | RiftLab