Cohere

Its product layer is North (an agentic workplace platform) and Compass (enterprise search/indexing); Cohere Labs is its open research arm, which publishes the Aya multilingual family and open weights for most Command models on Hugging Face. As of the September 2026 pricing page Cohere has visibly pivoted its public pricing away from broad pay-as-you-go token pricing toward free/open weights plus enterprise contracts and Model Vault instance pricing.

A Access and limits

Sign up at dashboard.cohere.com; a trial API key is created automatically. Trial keys are free but capped at 1,000 API calls/month and are not licensed for production/commercial use. Production keys require Owner privileges and completion of the "Go to Production" workflow on the Billing and Usage page; production usage is pay-as-you-go, billed monthly or when the outstanding balance reaches $250.

IMPORTANT — as of the live pricing page (2026-09), production keys behave like trial keys for the newer Chat model variants (Command A+, Command A Reasoning, Command A Translate, Command A Vision), i.e. 20 req/min and 1,000 calls/month, and production use of those requires contacting sales@cohere.com. Published per-token production rate limits (500 req/min) apply only to Command A, Command R+, Command R, Command R7B and North Mini Code.

Public per-token list prices are now published ONLY for Command R ($0.15/$0.60 per 1M), Command R7B ($0.0375/$0.15 per 1M) and Embed 4 ($0.12/1M text tokens, $0.47/1M image tokens); Rerank 4 Fast is $2.00 and Rerank 4 Pro $2.50 per 1,000 searches (a search = one query against up to 100 documents); Parse 5 is $1.50 per 1,000 pages. Command A+, North Mini Code and Cohere Transcribe are listed as free (free API key access, free model download). Legacy/"existing customers" prices still on the pricing FAQ: Command $1.00/$2.00, Command-light $0.30/$0.60, Command R 03-2024 $0.50/$1.50, Command R+ 04-2024 $3.00/$15.00, Command R+ 08-2024 $2.50/$10.00, Aya Expanse (8B and 32B) $0.50/$1.50 per 1M.

Cohere no longer publishes list prices for Command A, Command A Reasoning/Translate/Vision, Rerank 3.5, the Rerank v3.0 models or the Embed v3 models — those are contact-sales. Dedicated capacity is sold via Model Vault at fixed hourly/monthly instance rates: Embed 4 Small $4.00/hr or $2,500/mo, Embed 4 Medium $5.00/hr or $3,250/mo, Rerank 3.5 Medium $5.00/hr or $3,250/mo, Rerank 4 Fast Medium $5.00/hr or $3,250/mo, Rerank 4 Pro Medium $5.00/hr or $3,250/mo and Large $10.00/hr or $6,500/mo, Parse 5 Medium $4.00/hr or $2,500/mo and XL $7.00/hr or $4,300/mo; Cohere Transcribe starts from $3.75/hr/instance. Other endpoint rate limits: Embed 2,000 inputs/min (trial and production), Embed images 5 inputs/min trial / 400 production, Audio Transcriptions 5 req/min trial and contact-sales for production.

No public batch-API discount and no published prompt-caching discount. Models are also resold through Amazon Bedrock, Amazon SageMaker, Microsoft Azure AI Foundry and Oracle OCI Generative AI, each with its own pricing. An OpenAI-SDK-compatible endpoint exists at https://api.cohere.ai/compatibility/v1.

Fine-tuning for command, command-light, command-r, classify and rerank was retired on 2025-09-15; Embed/Classify fine-tuning remains available. Retired and therefore excluded here: c4ai-aya-expanse-8b, c4ai-aya-vision-8b, embed-english-v2.0, embed-english-light-v2.0 and embed-multilingual-v2.0 (all retired 2026-04-04), and rerank-english-v2.0 / rerank-multilingual-v2.0 (shut down 2025-04-30).

C Calling the API

Cohere accepts requests in the OpenAI Chat Completions format, so any OpenAI-compatible client works by changing the base URL. Example uses command-r7b-12-2024.

curl https://api.cohere.com/v2/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "command-r7b-12-2024",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

from openai import OpenAI

client = OpenAI(
    base_url="https://api.cohere.com/v2",
    api_key=os.environ["API_KEY"],
)

response = client.chat.completions.create(
    model="command-r7b-12-2024",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

A compatibility layer is not always the provider's full API — caching, strict tool schemas and structured output are commonly unsupported there. Check the official docs before relying on an advanced feature.

D Official references

• Homepage
• API documentation
• Pricing