Mixtral 8x22B: April 2024 Sparse MoE and a Retired Hosted API. 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 released Mixtral 8x22B on April 17, 2024 as an Apache-2.0 mixture-of-experts model with about 141 billion total parameters and 39 billion active. The published release lists a 64K context and multilingual strengths, including mathematical and coding tasks. A later model lifecycle page says the managed model is retired, with March 30, 2025 as retirement date; that must not be confused with the availability of previously downloaded licensed checkpoints.
A useful application and evaluation scenario
Before reusing an archived Mixtral installation, check tensor precision, memory topology, chat template and whether expert routing is efficiently supported by the inference stack. One useful test is a multilingual technical manual search with a human-labeled document set, testing both retrieval coverage and answer attribution. Observe throughput under several concurrent requests because sparse architecture can distribute memory and compute unevenly.
Limits, risks and uncertainties
It is incorrect to calculate model memory from only 39B active parameters. All experts typically require accessible weight storage even though a smaller subset runs per token. The self-reported model comparison does not establish an independent rank; the 2024 feature list also cannot prove the API is still live in late 2026.
Alternatives and the decision rule
Choose Mixtral for reproducibility or a validated self-hosted use case with a compatible license. For a fresh hosted integration, prefer an actively supported Mistral model and follow its separate release and price records. Leave the retired API string unchanged in any historical dataset rather than remapping it to a newer offering.
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.