Cohere Command

Command is the model that gave Cohere's whole product family its name. It arrived on 8 November 2022 as a beta called command-xlarge-20221108, described by the company as a generative model conditioned to respond well to single-statement commands - at a moment when instruction-following was still a novelty and ChatGPT was three weeks away from launch. The model was rebuilt continuously rather than versioned. In January 2023 Cohere replaced the dated snapshot with command-xlarge-nightly, retrained every week, and added a smaller sibling. When the company standardised its naming later that year, command-xlarge became simply command and command-medium became command-light. Users calling the old names were redirected without changes to their code. By the standards of what followed it was a small machine: a 4,000-token context window and a 4,000-token output limit, text only, no tool calling and no citations. It answered through the Chat endpoint and the now-retired Summarize endpoint, and it was the last Cohere generation whose parameter count the company never disclosed. Fine-tuning was available until 2025, then withdrawn along with the model. Cohere deprecated Command on 15 September 2025, in the same announcement that took out Command R, Command R+, Command-light, the Summarize and Classify endpoints and the entire fine-tuning stack. Under Cohere's own lifecycle rules a deprecated model stays available to existing users until a shutdown date is assigned - and as of August 2026 no shutdown date has been announced. Its published legacy price, one dollar per million input tokens and two per million output, is roughly six times what the same company charges for Command R 08-2024, a newer model with a context window thirty-two times larger.

▸News

Cohere's oldest model costs six times more than its newer replacement — and has no shutdown date to force anyone off it

8/22/2026 · Cohere documentation and pricing page

▸Videos

No videos yet.

▸Reviews