FIELD GUIDE / MISTRAL

Mistral Devstral 2 Pricing: Why a Deprecated Coding SKU Needs a Historical Ledger

The question is not which Devstral 2 plan is cheapest today: Mistral documents the model as deprecated. A defensible cost comparison starts with its original model ID, discontinuation milestones and the price evidence available for.

The question is not which Devstral 2 plan is cheapest today: Mistral documents the model as deprecated. A defensible cost comparison starts with its original model ID, discontinuation milestones and the price evidence available for the period being audited.

Verification scope: Official sources were checked for the 25 October 2026 cutoff. List tariffs and rights apply only to the documented provider, region, tier, exact model and sales channel. This is an editorial draft, not a verified account quote; unknown effective dates remain null.

The model version is a billing identity

Mistral documents Devstral 2 as version 25.12 with model identifier devstral-2512 and a 256k-token context window. Its model card is dated December 9, 2025 and explicitly flags a May 22, 2026 deprecation date. That history matters for engineering accounting: a 2025 invoice for a coding agent should not be joined to a 2026 Mistral Medium 3.5 rate just because Medium later replaced Devstral. Archive the request date, literal API model ID, input/output meters, any cached or batch SKU, and the original billing channel. Without an archived official price for the exact transaction period, the numeric Devstral rate remains unknown.

Deprecation and retirement are separate operational events

Deprecation is a signal not to begin new integrations; retirement means an endpoint is no longer callable. Mistral’s model list records lifecycle milestones that can differ from the prominent date on a model card. Confirm both against the version registry when explaining historical spend or forecasting the effect of migration. For a long-running repository agent, a model swap also changes tool-call behavior, context allocation, failure rate and token consumption; historical unit prices alone cannot tell you which option would have been cheaper in production.

A migration comparison requires workload-level totals

Suppose an internal coding assistant runs 40,000 requests a month. To compare an earlier Devstral deployment with a newer replacement, collect fresh input tokens, cached tokens, output tokens, retry counts and time spent reviewing code for each route. The illustrative volume of 40,000 is not a claim about either model’s performance. Apply each historical rate only if an independently dated tariff supports it; otherwise present an unpriced component. Include tests, security review, CI minutes, human sign-off and failed agent actions as separate costs, rather than treating model tokens as the complete development bill.

Choosing a successor is a new procurement decision

Mistral directs users toward Mistral Medium 3.5 for new integrations. That is the vendor’s recommendation, not proof that the alternative will succeed on any particular repository. A team should run reproducible tests on representative issues, record security boundaries, and compare how models recover from tool errors. The AI 06 team owns capability and developer-tool analysis; this guide preserves the historical price ledger and avoids asserting a benchmark win or a current Devstral purchase option.

Editorial decision at the October 2026 cutoff

Because the original Devstral tariff and contract conditions were not independently recovered for every archival period, its price is recorded as null and not zero. A present-day Mistral pricing page is insufficient to reconstruct every invoice from 2025 or early 2026. Do not fill that missing historical rate with a replacement-model price or a reseller estimate. The authoritative next step is archived vendor documentation or the organization’s own dated invoice, coupled with the exact service SKU.

Decision checklist and limitations

Before purchase or deployment, retain the literal model or plan identifier, payer/merchant, country or cloud region, billing currency, exact usage meter, source URL, checked date, effective date if explicitly supported, and any plan commitment or promotional eligibility. Recalculate from the companion price and quota snapshots rather than assuming that a remembered list tariff will remain unchanged. A credit allowance, maximum input length, published token rate and confirmed billed currency answer different questions; never add them as if they were the same quantity.

When comparing another provider or route, use the same input/output distribution and acceptable result definition. Include unsuccessful attempts, optional tools, storage, tax, payment and human review only when those components are actually measured or supported by an explicit source. Vendor marketing and documentation are not independent product tests. This article makes no claim to have personally tested paid endpoints, purchased plans, or checked account-specific taxes or availability.

Sources & Further Reading

CONTINUE EXPLORINGExplore related files ↗