Which API Providers Can Run a Paid Production Trial of DeepSeek V4 Pro 0813 With Support Terms Agreed Upfront?

Legal and procurement can read the whole contract set for a paid DeepSeek V4 Pro 0813 trial before the first request: the Service Level Agreement, Services for Proof of Concept Agreement, Services Agreement and Console Terms of Service are published in full, and the commitments specific to the trial go into a signed Order or Form. On the technical side, the model has its own card in the model library on GMI Cloud MaaS, with a 20% off badge, and MaaS serves it through one OpenAI-compatible endpoint. A client built for that API points at the GMI base URL with a GMI key and selects the model by ID.

Under both the Services Agreement and the Proof of Concept Agreement, the specific terms of the signed Order or Form govern when they differ from the general agreement. Support response times, escalation contacts and the service level that applies to per-token usage therefore belong in that document.

What to agree before the first request

The published agreements, read for a trial

Service Level Agreement

Our Service Level Agreement sets service level objectives as monthly uptime percentages: at least 99.9% for instances in multiple regions and at least 99% for instances in a single region. We commit to commercially reasonable endeavors to meet them. If an objective is missed and the client is otherwise in good standing and meeting its obligations, the client can claim a service credit. A credit claim must be received within sixty days after the end of the month in which the event occurred, and a service credit must be used within two months of being made available.

For a per-token API trial, the MaaS page lists "Guaranteed SLAs with uptime and performance commitments" and the developers page shows a 99.9% uptime target. The Order is the place to record the uptime and performance commitments that apply to your DeepSeek V4 Pro 0813 traffic, in line with the per-client customization across pricing, policies and deployment that the MaaS page lists.

Services for Proof of Concept Agreement

Our Services for Proof of Concept Agreement covers fixed-term proof-of-concept services described in a signed Form. The client holds a license to use those services from the Commencement Date to the End Date. Fees are paid in full before the Commencement Date, and the client attends a brief training session before the POC begins.

The Form is where the trial's specific terms are written, and they govern where they differ from the agreement. Service levels and support for the POC are among the terms to set out there.

Services Agreement

The Services Agreement holds our general terms for services bought under a mutually executed Order, and the Order's specific terms govern on any conflict. Each Order has an Initial Term and renews for the periods the Order sets, unless the client gives written notice at least sixty days before the renewal. The Proof of Concept Agreement ties access to the dates in the Form. For a trial that is meant to convert, state the Initial Term, the renewal period and the notice date in the Order, so the trial ends on a date both sides chose.

Console Terms of Service

For a self-serve start, the Console Terms of Service apply. The client prepays by adding funds to an account balance, or we may issue an invoice that is payable within seven days of the invoice date. Prompt logging is an opt-in setting in the account, and the logging license in the terms applies only if the client has opted in.

If you want the trial's terms signed rather than accepted in the console, a trial with fixed dates and a decision at the end fits the Proof of Concept Agreement and its Form, and a trial that rolls straight into ongoing service fits the Services Agreement and its Order.

Rate limits and tiers

Rate limits are enforced per organization, as tokens per minute for LLM models, and a new organization starts at Tier 1. As credit is purchased, the organization is upgraded to the next usage tier automatically, and the rate limits page shows the upgrade applied within 24 hours. Its upgrade table sets the cumulative purchase that triggers each tier. A separate quota table lists a tokens-per-minute limit for each tier from Tier 1 to Tier 5, and the limit that applies to the DeepSeek V4 Pro 0813 trial is one more figure to record in the Order. Voucher redemptions do not count toward purchase.

Estimate the peak tokens per minute your trial traffic will reach. If Tiers 2 through 4 cover it, fund the account to that tier and allow for the 24-hour upgrade window before the start date. For a manual tier upgrade, including a higher tier, the page names [email protected].

Seeing what the trial costs and how it behaves

Create a dedicated API key for the trial. The Serverless Usage view filters by timezone, time range, daily or hourly granularity, API key and model name, so the trial gets its own report separate from other work on the account.

Usage filter fields: choose the trial key and the DeepSeek-V4-Pro-0813 model, and the report covers only that trial.

In the LLM API reference, a complete response carries a usage object with prompt_tokens, and streaming responses include usage in the final chunk, so your own logs can be reconciled against the usage report. The reference also lists context_length_exceeded_behavior, which defaults to truncate and accepts error. Include your full parameter set in the trial's first requests, so the usage report reflects the request shape production will use.

Data handling

Sensitive workloads can run on a zero-retention configuration, one of the options MaaS lists. If the trial carries sensitive data, name the zero-retention configuration in the Order and list the keys it covers. Confidentiality is a separate protection: the Proof of Concept Agreement and the Services Agreement each include confidentiality terms.

From trial to contract

Write the success criteria into the Order alongside the support terms, before day one. Use measures your team can read from its own records: error rate and latency from your logs, hourly token totals from the usage report, peak tokens per minute from your own request logs, and the time it took to get an answer on each support request you raise during the trial.

The conversion path is part of the same document. MaaS bills centrally across models, and the per-client customization covered above is part of going from demos to production. If sustained volume later calls for fixed latency, the developers page lists switching to a dedicated endpoint as the Scale step for models that support dedicated deployment, and our MaaS FAQ describes moving from MaaS to dedicated endpoints without changing APIs.

FAQ

Where do the trial's model and price conditions get written down?

The signed Order or Form is where both conditions are written for the trial: the DeepSeek-V4-Pro-0813 card named as the model, and the pricing the trial runs on. Self-serve console usage, by comparison, deducts from account credits at the rates on the pricing page.

If the trial runs under an Order rather than console credit, how is it invoiced?

Under the Services Agreement, we invoice monthly in advance for the fees set out in the Order, and payment in US dollars is due within ten calendar days of the invoice date, or by another form of payment both sides agree. To dispute an invoiced amount, send a written notice with reasonable supporting documentation within ten calendar days of the date of the invoice on which that amount first appears. The Order or Form states which payment route the trial uses.

Trial steps from model card to end date

1. Find the "DeepSeek-V4-Pro-0813" card in the model library, not the separate "DeepSeek-V4-Pro" card, and copy its model ID.
2. Agree the Order or Form: scope, dates, service level, support terms, data handling, renewal.
3. Fund the account to the tier your peak traffic needs, allowing 24 hours for a purchase-triggered upgrade to Tiers 2 through 4, or request a manual upgrade by email.
4. Create the trial's API key and point your OpenAI-compatible client at the GMI endpoint with the new model ID.
5. Run the workload with your full parameter set, and track usage by key and model.
6. Review against the criteria in the Order, then convert, extend or end the trial on the dates it names.

The DeepSeek-V4-Pro-0813 card and trial key creation are both in the GMI Cloud console, and our team can put the trial's terms into an Order.