Cohere Pricing 2026 — API Cost Calculator + Comparison

Where are you coming from?

How Cohere stacks up

Cross-vendor synthesis — what you can’t get from Cohere’s own marketing.

In the 2026 LLM landscape, Cohere (slug: cohere) distinguishes itself by rejecting the per-token consumption model favored by most competitors. While mistral (slug: mistral) offers a granular pricing structure for its flagship mistral-large-3 ($2/M input, $6/M output), Cohere provides a subscription-based approach. This makes Cohere an outlier in the llm-api category, where variable costs are the standard. For organizations with high-volume output requirements, the lack of a $6/M output fee can result in significant savings, provided the seat cost is managed effectively.

Compared to specialized vendors like voyage (slug: voyage), which focuses on embeddings, Cohere offers a broader suite of generative capabilities within its subscription. However, the choice between Cohere and a token-based vendor like mistral often comes down to the specific integration archetype. In a rag-knowledge-base scenario where the LLM is only 25% of the total TCO, the pricing model's impact is less pronounced than in an office-productivity-rollout where the LLM accounts for 80% of the total cost. In the latter case, Cohere's seat-based predictability offers a clear advantage for corporate budgeting and internal chargebacks.

How the pricing actually works

Tier structure, batch discounts, caching, mechanism details.

Cohere's pricing mechanics are centered on seats and subscriptions, a departure from the per-token model used by peers like mistral. This means that instead of paying for every million tokens processed by a model like mistral-small-4 ($0.1/M input, $0.3/M output), users pay a fixed fee for access. This model eliminates the need for complex token-tracking infrastructure and provides a 'ceiling' on monthly expenses, which is particularly beneficial for experimental or high-throughput workflows.

Because there is no per-token API pricing, traditional optimization strategies like prompt caching or batch API discounts—which are critical for managing costs with vendors like mistral—do not apply to Cohere in the same way. Instead, the focus shifts to seat utilization and tier management. Organizations must evaluate their usage against the subscription cost to determine the break-even point compared to a consumption-based model. For example, in an inference-only-chatbot workflow where the LLM is 95% of the TCO, the seat cost becomes the near-total driver of the project's financial viability.

What's changed recently

Last 30 days of price + plan movement.

No notable price movements in last 30 days. Pricing has been stable.

Watch out for

Gotchas, traps, and recent shifts that surprise buyers.

_NARRATIVE_PENDING_
_NARRATIVE_PENDING_
_NARRATIVE_PENDING_

Vendor comparison

Flagship + cheapest tier across 3 vendors. Cohere highlighted.

Vendor	Flagship model	Input / output	Cheapest model	Recent changes (30d)
Cohere	`command-r-plus`	$2.5/M in · $10/M out	`command-r` $0.15 / $0.6	stable
Mistral	`mistral-large-3`	$2/M in · $6/M out	`mistral-small-4` $0.1 / $0.3	stable
Voyage AI	—	—	—	stable

Who wins for what

7 common scenarios — best vendor for each.

Predictable monthly cost for high-volume internal chat

Winner: cohere · cohere
Cohere uses a seat-based subscription model rather than variable per-token pricing.
Lowest cost for flagship-grade API tokens

Winner: mistral · mistral-large-3
Mistral Large 3 offers a transparent $2/M input and $6/M output rate.
Budget-conscious small model deployment

Winner: mistral · mistral-small-4
Mistral Small 4 provides a very low entry point at $0.1/M input tokens.
Specialized RAG embedding performance

Winner: voyage · voyage
Voyage AI is the designated peer for embedding-specific workflows in this category.
Capping financial risk for experimental 'vibe coding'

Winner: cohere · cohere
Seat-based pricing prevents unexpected bill spikes from high token consumption.
Enterprise-wide search with fixed budgeting

Winner: cohere · cohere
The seat-based model is ideal for rag-knowledge-base workflows where LLM costs are 25% of TCO and need to be capped.
High-precision tasks requiring Mistral's flagship

Winner: mistral · mistral-large-3
Mistral Large 3 provides a clear per-token price of $2/M input and $6/M output for high-performance needs.

Integration & TCO context

The seat fee is one line item. These archetypes show full TCO with engineering + observability + compliance.

Inference-only Chatbot (no retrieval) LLM is ~95% of total TCO

Workflow: general-q-and-a · Fit for: vibe coder, smb
Solo developer with ChatGPT Plus + Claude Pro = $40/mo. Total monthly cost is ~$40 because there are no integration costs.

Implementation: ~1 eng-weeks initial + ~2 hrs/month ongoing
RAG Knowledge Base / Internal Q&A LLM is ~25% of total TCO

Workflow: enterprise-search · Fit for: smb, enterprise
SMB support RAG: $400/mo LLM tokens, $1500/mo total TCO including eng + observability + eval.

Implementation: ~4 eng-weeks initial + ~12 hrs/month ongoing
Code Agent Deployment (Cursor / Copilot at team scale) LLM is ~70% of total TCO

Workflow: developer-productivity · Fit for: developer, smb, enterprise
50-dev team on Copilot Business = $950/mo seats + $200/mo overage + $1500/mo eng oversight = $2650 actual.

Implementation: ~2 eng-weeks initial + ~6 hrs/month ongoing
Customer Support Agent (stateful, multi-channel) LLM is ~30% of total TCO

Workflow: customer-service · Fit for: smb, enterprise
SMB with 10K tickets/mo: $800 agent runtime + $2500 eng + $400 platform = ~$3700/mo.

Implementation: ~8 eng-weeks initial + ~24 hrs/month ongoing
Voice Agent (Call Center / Receptionist) LLM is ~35% of total TCO

Workflow: voice-customer-service · Fit for: smb, enterprise
Restaurant chain with 5K calls/mo on Gemini Live: $25 voice + $300 LLM + $4000 eng/observability = ~$4300.

Implementation: ~6 eng-weeks initial + ~16 hrs/month ongoing
Multi-tool Autonomous Agent (research / sales / ops) LLM is ~20% of total TCO

Workflow: agentic-automation · Fit for: enterprise
Fortune 1000 with research agent: $2500 LLM + $1500 platform + $12K eng = ~$16K/mo for ONE agent in production.

Implementation: ~12 eng-weeks initial + ~40 hrs/month ongoing
Self-hosted OSS LLM (vLLM / Ollama / TensorRT) LLM is ~50% of total TCO

Workflow: data-sovereignty · Fit for: enterprise, developer
Healthcare OSS deployment: $4500/mo H100 rental + $12K eng = $16.5K/mo. Break-even vs Claude Sonnet around 100M tokens/month.

Implementation: ~6 eng-weeks initial + ~60 hrs/month ongoing
Office Productivity Rollout (Copilot org-wide) LLM is ~80% of total TCO

Workflow: workforce-enablement · Fit for: smb, enterprise
500-seat enterprise on M365 Copilot: $15K/mo seats + $700/mo overage + $700 governance = $16.4K/mo.

Continue your research

Cohere for other audiences

Cohere for Vibe Coders Cohere for Solopreneurs & SMBs Cohere for Developers Cohere for IT Buyers Cohere for Enterprise

Head-to-head comparisons

🆚 Cohere vs Mistral 🆚 Cohere vs Voyage

Alternative vendors

→ Mistral pricing guide → Voyage AI pricing guide

Cost optimization

💰 How to cut your Cohere bill

Calculators

🧮 cost calculator Estimate Cohere bill at your token volume 🧮 cheapest model Compare Cohere models against all 12 vendors 🧮 token estimator Convert documents/conversations to token counts before pricing 🧮 prompt cache roi Estimate savings from Cohere's prompt caching 🧮 batch vs realtime Cohere batch API typically 50% off realtime 🧮 context window cost See how long-context tier affects cost

📊 Raw data appendix (pricing tables, all models, all sources)

Current API Pricing

Per 1M tokens, USD. Refreshed nightly from Cohere's pricing pages.

Last refreshed 2026-05-02 from vendor pages

LLM / Chat Models

Model	Input $/1M tok	Output $/1M tok	Cached $/1M tok	Batch In $/1M tok	Batch Out $/1M tok	Context	Max Out	Modalities	Tags
Command R+ ⓘ	$2.50	$10.00	—	—	—	128K	4K	text	rag-specialist enterprise
Command R ⓘ	$0.15	$0.60	—	—	—	128K	4K	text	cheap rag-specialist

Embedding Models

Model	Input $/1M tok	Context	Dimensions	Tags
embed-english-v3.0 ⓘ	$0.10	—	1024	rag-specialist
embed-multilingual-v3.0 ⓘ	$0.10	—	1024	multilingual rag-specialist

🧮 Estimate your monthly bill → Compare against all 12 vendors →

Recent Price Movements

Changes detected by our crawler in the last 30 days

✓

No price changes detected in the last 30 days. Pricing has been stable.

Cheapest Cohere Model by Use Case

Picked algorithmically from current pricing — refreshes when prices move.

High-volume classification / labeling (low quality bar)

Pick Command R

$0.15/M in $0.60/M out

Lowest output cost in Cohere family at $0.60/M tokens.

Hard reasoning / complex agentic / code generation

Pick Command R+

$2.50/M in $10.00/M out

Premium tier in family; Cohere's strongest reasoning/coding model.

Pricing Mechanism Facts

Cache rates, batch discounts, SLAs — every claim cited verbatim from vendor docs.

vendor published Cohere — Pricing 2026-04-24T04:00:00.000Z

Cohere Embed v4 pricing is billed per instance with hourly and monthly rates based on performance tier.

**Embed 4** Small $4.00 $2,500 **Embed 4** Medium $5.00 $3,250 — Cohere — Pricing

Rates shown are per instance; billing can be calculated hourly or through longer-term commitments (monthly or annual).

vendor published Cohere — Pricing 2026-04-24T04:00:00.000Z

Cohere Rerank v3 pricing is defined per search unit, where one search unit equals one query with up to 100 documents to be ranked.

A single search unit is defined as one query with up to 100 documents to be ranked. — Cohere — Pricing

If any document exceeds 500 tokens (including the length of the search query), it is automatically split into multiple chunks. Each chunk is treated as an individual document and counts toward the total number of documents ranked for that search.

Cohere vs Mistral — Per-1M-Token Cost

Flagship-tier output tokens, standard pricing (no batch, no cache).

Cohere

Command R

Input $0.15 /1M

Output $0.60 /1M

128K context

Mistral

Mistral Small 3 (deprecated)

Input $0.20 /1M