Does Cohere have an SLA? The 99.5% uptime objective, its 10/20/30% credits, and the 30-day clock that starts before the month ends

When an LLM API goes down, the product built on it stops thinking: support bots stall, summaries come back empty, embedding jobs back up in the queue. Afterwards the platform or FinOps lead asks the usual question — *does Cohere owe us anything?* Unlike many AI vendors, Cohere publishes an answer, with a credit table. Its claim clock is unusual, though, and the table has two gaps.

Does Cohere have an SLA with service credits?

Yes. Cohere publishes a Service Level Objective (SLO) for the Cohere API SaaS Service at cohere.com/slo. It sets a 99.5% Monthly Uptime Percentage for each Covered Service. If a calendar month falls short, Cohere credits 10%, 20% or 30% of the monthly invoice for the Covered Service that missed, applied to your next monthly invoice.

To get it, you email support@cohere.com within 30 days of the time you become entitled to the credit. That clock starts when the month's uptime drops below 99.5%, which can happen mid-month, so waiting for the month to close can cost you days. Cohere then decides, in its sole discretion and using its own internal logs as well as yours, whether the credit is owed.

What does the Cohere SLO actually promise?

Source: Cohere's published Service Level Objective, which carries no effective or last-updated date and was read on 9 October 2026. Ontracko's Cohere SLA profile has the same tariff.

One wording note: the HTML table on cohere.com/slo lists "dashboard.cohere.ai" where the downloadable SLO document says "os.cohere.ai". We transcribe the document. If your claim concerns the dashboard, quote the document's wording and name the host you actually used.

The word "objective" matters less than it might seem. The figure sits inside a "commercially reasonable efforts" wrapper, but the remedy is triggered by the number itself: if Cohere fails to meet the SLO in a calendar month, the customer is entitled to the credits. That makes it a priced commitment, not just a stated target.

How much downtime does Cohere's 99.5% allow?

In a 30-day month (43,200 minutes), 99.5% allows 216 minutes, or 3.6 hours, of downtime per Covered Service. In a 31-day month it allows about 3.7 hours. Beyond that, the table prices the shortfall like this:

The two "not priced" rows are Cohere's own boundaries, not ours. The 20% band ends at 97.99% and the 10% band starts at 98%. The 30% band is "<94.99%" and the 20% band starts at 95%. A month that lands in either sliver is below the 99.5% SLO, but the table sets no percentage for it. Each sliver is only a few minutes wide in a 30-day month, so it will rarely matter. If your month does land there, name the interval in the claim rather than assuming a tier. For the general arithmetic, see what a 99.9% SLA actually allows and how SLA uptime is measured.

When does Cohere's 30-day claim clock start?

When you become entitled to the credit, not at month-end. The SLO says you must notify Cohere at support@cohere.com within 30 days of the time you become entitled to receive a Service Level Credit. Most cloud SLAs count from the end of the month in which the incident happened. Here the safest reading is that the clock can start mid-month, as soon as cumulative downtime pushes the month below 99.5%. Missing it forfeits the credit.

There is a decoy to avoid: the Cohere SaaS Agreement has a 45-day "Disputed Charges" clause. That is a billing-dispute clock, not the credit window. The SLO's 30 days is the one that applies to downtime. Our SLA claim deadlines by vendor roundup compares this with other vendors.

What doesn't count as Cohere downtime?

The SLO excludes:

• Features marked Alpha or Beta, or excluded from the Covered Services in Cohere's documentation.
• Unavailability caused by your own (or your Permitted Users') acts or omissions, modifications, or breach of the Agreement.
• Not following Cohere's recommendations, including its minimum system requirements.
• Cohere acting at your request or direction.
• Demand spikes you drove that were not agreed in writing beforehand.
• Scheduled or emergency maintenance.
• Downtime of third-party service providers.
• Force majeure.
• Suspension permitted under the Agreement.

The demand-spike exclusion matters for AI workloads in particular. If a batch embedding backfill you launched is what saturated your capacity, Cohere can exclude those minutes. Throttling under a load you agreed in writing is a different matter.

How to claim a Cohere SLA credit

1. Log every incident against the affected Covered Service (api.cohere.ai or os.cohere.ai) with start and end times from your own monitoring. Use the Cohere status page to corroborate.
2. Keep a running total of each service's downtime for the month. The moment it passes 216 minutes in a 30-day month, the month has missed 99.5% and your 30-day clock is running.
3. Remove excluded time: Alpha/Beta features, maintenance, your own changes or unagreed demand spikes, third-party provider outages, and force majeure.
4. Work out the Monthly Uptime Percentage and match it to the table. Check whether it lands in one of the two unpriced gaps.
5. Email support@cohere.com within 30 days of becoming entitled. Include the date and time of the SLO failure, the affected Covered Service, your downtime evidence, and the tier you are claiming. Keep the sent timestamp.
6. Answer any follow-up questions from Cohere promptly. The SLO lets Cohere ask for more information, and failing to provide it forfeits the credit.
7. Check that the credit appears on your next monthly invoice. If you are leaving Cohere, file before the Agreement terminates or expires, because unused credits are forfeited then.

A worked example

Say your api.cohere.ai invoice is $5,000 a month. In a 30-day month, the service was unavailable for a cumulative 5 hours (300 minutes) of non-excluded downtime. Uptime is (43,200 − 300) ÷ 43,200 ≈ 99.31%, which falls in "98% to < 99.5%". The credit is 10% × $5,000 = $500.

If the month's total had been 15 hours (900 minutes), uptime would be (43,200 − 900) ÷ 43,200 ≈ 97.92%. That falls in "95% to 97.99%", so the credit is 20% × $5,000 = $1,000. If instead the outages added up to only 3 hours, uptime would be about 99.58%: the SLO is met and nothing is owed. The credit calculator on the Cohere SLA page runs this table against your own invoice, and our guide to calculating any SLA credit explains the method.

Frequently asked questions

What is Cohere's uptime guarantee?

Cohere's published SLO sets a 99.5% Monthly Uptime Percentage for each Covered Service (os.cohere.ai and api.cohere.ai). Missing it earns a credit of 10%, 20% or 30% of that service's monthly invoice.

Am I owed a credit if Cohere was down for two hours?

Not on its own. Two hours in a 30-day month is about 99.72% uptime, which is above the 99.5% objective. A credit is owed only once a Covered Service's non-excluded downtime for the month passes 3.6 hours (216 minutes).

How long do I have to claim a Cohere credit?

30 days from the time you become entitled to the credit, by email to support@cohere.com. To be safe, assume the clock starts when the month's uptime first drops below 99.5%, not when the month ends.

Does Cohere refund money for outages?

No. Credits go on the next monthly invoice and apply only to outstanding or future invoices. Cohere is not required to issue refunds or make payments against credits, and credits are forfeited when the Agreement terminates or expires.

Is the credit a share of my whole Cohere bill?

No. It is a percentage of the monthly invoice for the Covered Service that missed the SLO. Price each service separately.

Who decides whether I get the credit?

Cohere does. It decides in its sole discretion, using the information you provide and its own internal system logs. That is why your own timestamped evidence matters.

How does Cohere compare with OpenAI and Anthropic?

See our OpenAI and Anthropic SLA guide for those vendors' terms. Cohere is one of the AI vendors in our catalogue that publishes a credit table openly. For the wider list, see which SaaS vendors actually pay SLA credits.

Methodology & caveats

Every term above is transcribed from Cohere's published "Service Level Objective (SLO) — Cohere API SaaS Service" at cohere.com/slo, read on 9 October 2026. That includes the Covered Services, the 99.5% objective and its "commercially reasonable efforts" wording, the calendar-month measurement, the 10/20/30% table with its exact boundaries, the 30-days-from-entitlement deadline, the email claim, the sole-discretion determination, the credit-form and forfeiture terms, the sole-remedy clause, and the exclusions. The document has no effective or last-updated date, and Cohere may amend it, so re-read it before you file. If your Order or agreement with Cohere states different terms, it controls.

The downtime minutes and the two gap widths are our own arithmetic on 30-day months. The note that the 45-day Disputed Charges clause is not the credit window comes from our reading of the Cohere SaaS Agreement. Ontracko scores status-page incident duration, which shows that an event happened but is not the Monthly Uptime Percentage Cohere computes, so a claim also needs your own logs. For more context, see why SLA credit claims get denied, the Pinecone SLA guide, the reliability rankings and the SLA glossary.

*With Cohere, the claim clock can start before the month ends. Ontracko watches the Cohere status feed for free, times each incident against the 99.5% objective, and alerts you when a month crosses into credit territory, so the email goes out well inside 30 days. Monitor Cohere free; we take 8% only on credits we recover. See Cohere live status or open the Cohere SLA page.*