Our take
Cohere is a focused API for the vendor's own command models, with a small catalogue and a clear speed-to-price trade-off across its three tiers. It does not train on prompts, but its compliance and geographic footprint are undocumented.
Use this when you need Cohere's own command models specifically, or for budget-conscious access to the entry-tier model which runs fastest and cheapest in the lineup. Skip it if you need a broad model catalogue, if you require verified SOC 2 or a known EU data-residency endpoint, or if your workload needs throughput that holds up on larger models.
- Fastest throughput on the cheapest model in the lineup — Command R7B runs at 75 tokens per second, three times faster than Command A and three times faster than Command R.
- Clear pricing separation by model capability: the entry model sits well below the mid-tier, which in turn sits far below the flagship.
- Privacy-safe prompt handling: the vendor does not train on prompts.
- Very small model catalogue — only three tracked offers, all from a single model family.
- Throughput drops sharply on larger models: Command R runs at 25 tokens per second, three times slower than Command R7B; Command A at 33 tokens per second is still more than twice as slow as R7B despite being the flagship.
- Flagship pricing is steep versus the entry model, and compliance and geographic footprint are both undocumented: SOC 2, EU endpoint and headquarters country are all unverified in our data.
Privacy & data handling
- ✓
- Yes
- ✗
- No
- No answer on record
The colour says whether the answer favours you, not whether it is a yes — not training on your prompts earns a green cross, no zero-retention option earns a red one.
Prompt retention: 30 days.