Mistral Medium 3: May 2025 Launch and the 2026 Deprecation Boundary. This reference distinguishes the original release from hosted endpoints, downloadable weights and subsequent version changes. It is written for people deciding whether an exact model or checkpoint is suitable for a real application, not for leaderboard marketing.
What the sources actually establish
Mistral introduced Medium 3 on May 7, 2025 for enterprise coding and multimodal understanding, with hybrid and private deployment paths described in its announcement. The provider’s model record identifies the API ID `mistral-medium-2505` and a 128K context, but marks that version deprecated on May 22, 2026 in favor of a later Medium release. An enterprise customer therefore needs to distinguish history from current onboarding recommendations.
A useful application and evaluation scenario
An architecture team can benchmark Medium 3 on a mix of code summaries, ticket routing and document analysis, keeping model-generated answers separate from retrieval output and OCR preprocessing. Test structured response validity and function-call argument quality, especially when migrating to a newer Medium generation. Include representative languages and policies, not just an English demonstration prompt.
Limits, risks and uncertainties
The vendor’s cost and performance positioning is not independently reproduced here. Hybrid and on-prem availability depends on agreement and infrastructure, not just the marketing claim. Deprecation of the provider route does not establish that the model was immediately disabled; check dated lifecycle documentation for actual retirement and use AI10 for any region-specific pricing.
Alternatives and the decision rule
Retain Medium 3’s exact ID for audits and test a supported successor for new production. If a legacy system is contractually pinned to the older API, create a sunset runbook with acceptance testing before migrating. Do not label the parent family name `mistral-medium-latest` as a fixed version.
A reproducible test before adoption
Prepare a small held-out set from data the organization is authorized to process. Include simple, ambiguous and deliberately unanswerable requests; specify the expected answer or evidence and a human-reviewed scoring rule. Record model identifier, endpoint and region, tokenizer or SDK version, prompt, tool settings, output budget and whether external retrieval was allowed. Do not present a provider-supplied benchmark as an independently measured result. Separate model mistakes from retrieval, OCR and application validation failures.
Operational checklist and dated evidence
A production transition must test 429 responses, timeouts, refusals, invalid structured output and model-unavailable errors. Keep a reversible rollout or safe fallback instead of silently substituting a model behind an old identifier. Measure latency and usage on the actual workload before planning budgets; join dated token quantities to the AI 10 price ledger. Verify enterprise residency, region, support entitlement and data retention with the actual contract, not a public release story.
Sources & Further Reading
- mistral.ai — documentation/reference (supplier-provided reference; not independently checked in this pass)
- docs.mistral.ai — documentation/reference (supplier-provided reference; not independently checked in this pass)
Source documentation establishes vendor-published facts, not independently measured model quality, tenant access, contractual rights or current prices. Use a dated pricing ledger and an account-specific entitlement check for deployment decisions.