Cohere Command R 08-2024

On 30 August 2024 Cohere refreshed both of its flagship text models and gave them timestamps: command-r-08-2024 and command-r-plus-08-2024. The Command R update is the more interesting of the two, because the model got smaller. The published weights hold 32.3 billion parameters against 34.98 billion in the March version — Cohere says the refresh halved the hardware needed to serve the model while delivering around 50 percent higher throughput and 20 percent lower latency. The savings were passed on. Input dropped from 0.50 to 0.15 dollars per million tokens and output from 1.50 to 0.60 — a 70 and 60 percent cut respectively, for a model Cohere describes as competitive with the previous, much larger Command R+. The functional changes are the ones an engineer notices. The model decides better whether a tool is needed at all, follows system-message instructions more closely, handles structured data both ways, and can now run retrieval workflows with citations switched off. It also declines unanswerable questions instead of inventing an answer. New in this release were safety modes — a safety_mode parameter with STRICT and CONTEXTUAL settings, letting a deployment tighten or loosen guardrails according to context, with core protections always on. The context window stayed at 128,000 tokens and the language coverage at ten optimised languages. The weights are on Hugging Face under CC BY-NC 4.0, so they are downloadable but not commercially deployable. Cohere still lists this model as live in August 2026, and recommended it as the migration target when the original Command R was deprecated in September 2025.

▸Videos

No videos yet.

▸Reviews