Command R+ Token Rates, Agent Loops & the Real Bill 2026 Guide
Command R+ is a usage-priced API at $3 per million input and $15 output, tuned for tool-calling and agentic work. There is no subscription, and the output rate plus agent loops decide the total.
Typical token rate
$3-$15/1M
input to output, output at five times input; usage-based, no subscription
Hidden fees
Yes
agentic loops generate heavily, dedicated endpoints quote-only, no discounted sandbox
Free tier
Trial only
a rate-limited developer trial for non-production testing, not a free plan
Cost transparency
Medium
scores 4 of 6 on our transparency checklist
Command R+ true cost, output-heavy
High· Verified July 15, 2026Command R+ is a usage-priced API with no subscription as of July 15, 2026. Input runs $3 per million tokens and output $15, a five-to-one ratio. A read-heavy retrieval call stays cheap, but agentic and tool-calling loops that generate a lot push the bill toward the output side. There is no free plan, only a developer trial for testing, and dedicated endpoints are quote-only. So the output rate and how much your agents generate decide the total.
- Input /1M$3
- Output /1M$15
- Output vs input5x
- Context window128k
- Free trialYes
- Monthly floor$0
- Dedicated endpointQuote
Command R+ has no monthly seat, so it sidesteps the $7.99 median across the 20 llm tools we track. It bills $3 per million input and $15 output, and agentic loops decide the total.
Command R+ savings on a usage API
Command R+ carries no seats and no subscription, so there is nothing to discount, and no student or nonprofit rate exists in July 2026. The free developer trial is your only no-cost route, and it suits testing rather than production traffic. Beyond that, every saving is engineering.
Since output is five times input, the biggest lever is generation. Keep agent responses tight, cap tool-use loops so a chain cannot run away, and reuse cached context where you can. Test on the trial's rate limits rather than iterating against production rates. At steady volume, a dedicated endpoint or a committed contract prices below the shared card, which is what the tactics below work through.
Cap the agent loops
Because output bills at $15 and agent chains generate a lot, capping the number of tool-use steps and trimming response length is the single biggest saving. A runaway loop is where an agentic Command R+ bill actually balloons.
Iterate on the trial, not production
There is no discounted sandbox, so tuning a workflow at full rates is a real cost. Doing as much iteration as possible within the trial's rate limits keeps development spend down before you move to production billing.
Commit volume for a lower rate
At steady, high volume a committed contract or dedicated endpoint prices below the shared per-token card. Reserved capacity only pays off when your traffic is consistent enough to keep it busy, so match it to real load.
Use the smallest capable model
Command R+ is powerful, but a simpler task may run cheaper on a lighter Cohere model or a rival. Since output dominates the bill, using the smallest capable model for each step lowers cost without hurting the result.
Negotiating a Command R+ usage contract
The published rate leaves nothing to negotiate, so the levers are all engineering: cap the loops, trim output, right-size the model. A real conversation opens only once your committed volume grows large enough for Cohere to price against, the point where dedicated endpoints and committed rates enter the picture.
Two plays do the heavy lifting, and each rests on the output rate being the figure that truly decides an agentic bill.
Commit volume for a rate under the card
- Target
- Committed-use contract
- Argument
- Guarantee a monthly token volume, weighted toward output, for a rate below the public $3 and $15. Since output dominates an agentic bill, negotiating the output rate specifically is worth more than shaving the input side.
Anchor on a cheaper agentic API
- Target
- Any high-volume deal
- Argument
- Amazon Nova near $0.035 per million and Gemini at $1.25 in undercut Command R+ sharply. If your workflow does not need its specific tool-calling tuning, make Cohere match the economics or defend the gap.
Move output-heavy agents to a gentler API
- Target
- Generation-heavy workloads
- Argument
- The five-to-one output ratio hurts most on chatty agents. If a rival with a three-to-one ratio like Mistral Large handles the task, the structural saving on output beats any rate you could negotiate here.
When a Command R+ deal is worth timing
Per-token spend offers no timing angle, since the rate card pays no mind to the calendar. A committed-use or dedicated-endpoint talk is the only thing worth timing, and it moves on the standard sales cycle. Cohere's team chases quarterly numbers, so a committed deal closed near a quarter end usually prices better than one opened at the start. The larger timing question, though, is your own workload, because agent behavior drives the bill more than the rate does.
Jan
Feb
Mar
Q-END
Apr
May
Jun
Q-END
Jul
Aug
Sep
Q-END
Oct
Nov
Dec
Q-END
Pro tip: Profile your agents before you negotiate. Because output drives the cost, a measured breakdown of how many tokens your chains generate is worth more at the table than a headline request volume, and it keeps the endpoint sized right.
Command R+ costs: what moves at volume
Point your asks where Cohere actually has room. The published token rate holds firm below committed volume; the give is in a committed output rate and a dedicated endpoint.
Usually negotiable
- Committed output token rateHIGH
- Dedicated endpoint pricingHIGH
- Rate limits at scaleMEDIUM
- Data-handling and retention termsMEDIUM
- Payment terms (Net 30/60)LOW
Rarely negotiable
- The published $3 input and $15 output rates
- The five-to-one output ratio
- The 128k context window
- The absence of a discounted development sandbox
Command R+ negotiation email generator
Fill the fields and the draft writes the copy for you, pulling current competitor token prices straight from our catalog. Route it to your Cohere account contact or the sales form. Lean the ask toward output, since that is where an agentic bill really sits. Give your monthly input and output volume, place a cheaper API rate beside it, request a committed output rate, and name a decision date.
guaranteed token spend, weighted to output, for a lower rate
Hi Command R+ team, I lead tooling decisions at [Your company], and we are evaluating Command R+ Team seats for a team of 10-50 people. As part of this evaluation we are also looking at Amazon Nova, which comes in at $0.035 per 1M input, and Google Gemini at $1.25 per 1M input. Can you help us understand the value difference at your current rates? We are ready to commit to an annual term. What is the best rate you can offer on annual billing, and can you cap the renewal price in the contract? We are aiming to sign before the end of this quarter, and budget sign-off is already in place. Could you share a proposal covering the per-seat or per-credit rate, the renewal terms, and any programs we qualify for? Best regards, [Your name] [Your company]
Send it Tuesday to Thursday, and follow up once after 3 business days.
Before you send
- Have your monthly output volume ready, since output is where an agentic Command R+ bill actually lands.
- Send midweek, because a note arriving Tuesday through Thursday clears faster than a Monday or Friday one.
- Do not lead with your budget. Let Cohere quote the committed rate first, then push on the output side.
- Name a cheaper agentic API by price. The generator inserts its current token rate into the copy for you.
- Ask for the output rate and any dedicated-endpoint terms in writing before you move production traffic.
- Follow up once after a few business days, then read continued quiet as a read on your position.
Command R+ cost mistakes in agent pipelines
Each of these comes from how output-weighted usage billing meets agentic workloads, and all are avoidable early.
Budgeting from input alone. Output costs five times more, so generation, not context, sets an agentic bill.
Leaving agent loops uncapped. A runaway chain generates unbounded output at $15 per million, the classic blowup.
Iterating against production rates. There is no discounted sandbox, so tune within the trial limits where you can.
Reserving an endpoint for bursty traffic. It bills whether busy or not, so it only pays off at steady load.
Running everything on the flagship. A lighter model may finish the same step in fewer tokens for far less.
Command R+ rivals for agentic and RAG work
Command R+ is capable but now an older agentic model, so leverage means pointing out what does the same tool-calling job for less. The three below come from our verified catalog. Switching is not the point. What you want is a benchmarked rival rate, measured on your own agent workload, so a committed-output talk with Cohere stands on numbers rather than its framing.
Amazon Nova
Nova Micro input, tool use on Bedrock
$0.035/1M
The price floor for agentic work. Nova undercuts Command R+ by orders of magnitude on input, so it is the anchor when your agents do not need its specific tuning.
Google Gemini
input rate, function calling built in
$1.25/1M
A cheaper frontier model with native function calling. The cross-shop when your agent pipeline could run on a general model rather than a retrieval-tuned one.
Mistral Large
input, gentle 3:1 output ratio
$2/1M
A gentler output ratio matters when output drives your bill. For generation-heavy agents, Mistral's three-to-one beats Command R+'s five-to-one directly.
Script“Our agents are output-heavy, and Command R+ is $15 out per million. Mistral Large is $6 out, Nova far less. What committed output rate keeps this workload with Cohere?”
Is Command R+ worth it? A usage-cost read
Command R+ is a solid choice for enterprise retrieval and agentic work that needs tool calling and multilingual support, and a poor one to run without watching the output meter. There is no flat floor, which is friendly for low-volume testing, but the cost tracks generation directly, and agentic loops generate a lot. It is also now the older tier below Cohere's Command A, so it is proven rather than the sharpest option.
The cost that surprises people is not input, it is output at five times the rate, multiplied by however much your agents produce. There is no discounted sandbox, so even development runs at production rates. So cap the loops, trim responses, right-size the model, and iterate within the trial limits before you ship.
Above real volume, negotiate the output rate specifically, since that is where an agentic bill lives, and price a cheaper rival's output rate against it. The full rates sit on the Command R+ pricing page. For many agent workloads a gentler output ratio elsewhere is the bigger saving, and the Command R+ alternatives page shows the rates to compare.
Command R+ pricing and discount FAQ
How does Command R+ charge?
+
Command R+ is a usage-priced API with no subscription. Input costs $3 per million tokens and output $15, a five-to-one ratio, across a 128k context window. You pay only for the tokens your requests consume, so a read-heavy retrieval call stays cheap while an agentic, tool-calling workflow that generates a lot leans on the output side. There is a free developer trial for non-production testing, not a free plan, and dedicated endpoints that bypass shared rate limits are quote-based.
Why does Command R+ get expensive with agents?
+
Because it is built for agentic, tool-calling work, and that work generates a lot of output at the $15 per million rate. A chained workflow reasons across several steps, and every step bills on the output side, so a talkative agent runs up cost that a simple per-request estimate misses. Usage pricing is hard to forecast when the token count depends on how the agent behaves. The fix is capping tool-use loops and trimming response length, which controls the output volume that drives the bill.
Is there a free Command R+ tier?
+
No, only a free developer trial. It gives rate-limited access for non-production testing, enough to prove a pipeline before you pay, but it is not a free tier for live traffic. There is no discounted sandbox beyond it either, so once you move past the trial limits, tuning a workflow runs at full production rates. Plan to do as much iteration as possible within the trial before shipping, because development itself becomes a billable cost on this model.
Does Command R+ offer student or nonprofit rates?
+
None published in July 2026, because a usage-priced API with no subscription gives a discount nothing to attach to. The savings are engineering: cap agent loops, trim output since it runs five times input, right-size the model, and iterate inside the trial limits instead of at production rates. At steady, high volume, a committed contract or dedicated endpoint prices under the shared card, and that is where a genuine rate reduction is actually won.
What is the difference between Command R+ and Command A?
+
Command A is Cohere's newer flagship, and Command R+ now sits below it as the proven older release. For cost, that matters mainly because a newer model may complete the same task in fewer tokens, which lowers the bill even at a similar or higher per-token rate. Command R+ remains capable for enterprise retrieval and agentic tool calling. Whether to move to Command A is a quality-versus-cost question worth testing on your own workload rather than assuming the newer tier is always cheaper in practice.
Are Command R+ dedicated endpoints cheaper?
+
Only at steady, high volume. A dedicated endpoint bypasses shared rate limits by putting you on reserved capacity, which you pay for whether or not it is busy. For consistent traffic that keeps it full, that can be cheaper per token than the shared card. For bursty or low-volume workloads, it leaves you paying for idle capacity, so it costs more. Profile your real traffic first, because reserved capacity only pays off when your load is predictable enough to keep it working.
How do you lower a Command R+ bill?
+
Focus on output, since it costs five times input and agentic loops generate a lot of it. Cap the number of tool-use steps, trim response length, and right-size the model so simple steps do not run on the flagship. Iterate within the trial's rate limits rather than at production rates, because there is no discounted sandbox. At scale, negotiate the output rate specifically in a committed contract. Those moves target the exact part of the bill that agentic work inflates.
Is Command R+ cheaper than a general model API?
+
Often not, especially for output-heavy agents. Command R+ is $3 in and $15 out per million, while Amazon Nova starts near $0.035 and Mistral Large runs $2 in and $6 out with a gentler three-to-one ratio. If your pipeline needs Command R+'s specific retrieval and tool-calling tuning, it may be worth the premium. If it does not, a general model with a lower output rate can be dramatically cheaper on an agentic workload. Benchmark the output cost on your own agents before committing.
Explore Command R+
Every page on Command R+ in one place, you are on cost guide.
Sources & verification
| Source | What was checked | Last checked |
|---|---|---|
| Command R+ official pricing | Verified plan prices, renewal rates and credit allowances | July 15, 2026 |
| Command R+ website | Official vendor website | July 15, 2026 |
| Command R+ pricing on ComparEdge | Current prices for every plan, with the cost calculator | July 15, 2026 |
Every fact on this Command R+ pricing page is tied to a named source and a verification date. Freshness-sensitive figures trace to the sources above; verify against the vendor before relying on them.