Open Model WeightsMISTRAL AI · FIELD VERIFIED
Primary sources last checked on 30 Oct 2026. Verification freshness is tracked separately from factual change, and unknown values remain explicit rather than inferred.
MODEL INTELLIGENCE · SOURCE-FIRST
Open Model Weights records Mistral Large 4 (1T-A52B) as a 1050B total / 52B active (publisher-declared)-parameter text and images → text open-weight model in the Mistral Large 4 family, with Mistral AI listed as the developer. The checked source repository provides 55 recognized weight artifacts, format evidence for Safetensors and a verified context window of 1,048,576 tokens. Values that are not evidenced in the checked sources remain explicitly undisclosed or unknown.
The repository declares Mistral Ecosystem Open-Weight License v26.10. Commercial use under Mistral Ecosystem Open-Weight License conditions; operating a paid Model-as-a-Service requires a separate commercial agreement with Mistral.. Open Model Weights reports the checked license evidence and classification; this is not legal advice.
No base model is declared in the checked standard lineage metadata, so Open Model Weights does not infer one. Training recipe status: training-described-in-model-card; data disclosure status: described-in-model-card.
+ Save to my shortlistMonitor changes →Local browser monitoring · no background email alerts
DIRECT ANSWER · FEATURED MODEL BRIEFING
Source-grounded editorial layer · reviewed 2026-10-30
Mistral Large 4 (1T-A52B) is Mistral AI's October 29, 2026 original downloadable multimodal mixture-of-experts checkpoint. Its official repository provides source-verifiable weights, a 1,048,576-token configuration and a custom Mistral Ecosystem Open-Weight License. Mistral describes roughly 1.05 trillion logical parameters, 52 billion active per token including embedding and output layers, and text plus image input. Open Model Weights classifies it as Leading because an independently operated open-weight intelligence index includes the released model, not because we independently ran inference or verified publisher leaderboard claims.
WHY IT MATTERS
The checkpoint connects an API-previewed frontier family to concrete downloadable original weights, an exact revision, a model card and a checkable license.
A one-million-token configuration is relevant to long documents and multimodal analysis, but represents a source-declared context capacity, not an independently measured production limit.
Independent comparative benchmark inclusion gives a basis for Leading discovery separate from manufacturer performance claims; Mistral's license has a material separate-agreement rule for MaaS operators.
MODEL-SELECTION VIEW
The October 6 Mistral Large 4 API preview and October 29 release of official open weights are different publication events. Our verified record points to the Mistral AI-owned 1T-A52B repository, its retained revision hash, exact file manifest and model configuration.
We preserve the distinction between the producer's architectural claims and properties extracted from the released artifact. Independent benchmark coverage is the reason for featuring the model among Leading selections; it is not an endorsement of any particular hosted endpoint or unverified provider comparison.
Mistral describes roughly 1.05T parameters overall and 52B activated per token including embedding and output layers, combining large stored capacity with sparse computational routing. Those are distinct notions: the active count does not imply a small checkpoint or memory footprint.
The first-party config contains a 1,048,576-token context setting. Real accuracy, latency, attention-memory cost and serving concurrency at that length require reproducible evaluation with a pinned stack; a configuration value is not a speed benchmark.
The Mistral Ecosystem Open-Weight License is a custom grant, not a synonym for Apache 2.0 or MIT. It contains conditions on redistribution and licensed use and requires separate authorization to operate a paid Model-as-a-Service business.
Teams should confirm whether a planned offering is an internal inference service, a SaaS feature or a paid standalone MaaS activity; the correct rights analysis depends on the actual license language and business model. Open Model Weights reports the source classification rather than giving legal advice.
Leading is an independent-relevance curation label: the model appears on a recognized external open-weight intelligence comparison, while the original release enables checkpoint-level analysis. Scores should be matched by test suite, date, high/low reasoning settings and model variant.
The label never means Open Model Weights measured accuracy, coding proficiency, cybersecurity, energy cost or agent success. Those would require separate controlled runs and independent evidence.
The checked 1T-A52B repository contains the original first-party artifact manifest, while optimized and low-bit distributions are separate outputs with distinct implementation assumptions. They may have different validation needs, weight-file geometry and runtime support. A performance claim made for a hosted version cannot automatically be ascribed to the original checkpoint.
For reproducible records, teams should preserve the exact repository revision, config, tokenizer, model-card version and a checkpoint-specific deployment script. A model's architecture and branded family name do not substitute for knowing which exact file hashes were evaluated.
The published million-token configuration is a source-checked capacity setting, not an assurance of equivalent retrieval precision, reliability or throughput over a million-token prompt. Validation should include a controlled corpus, a fixed set of questions, long-context distractor tests, multimodal image-text tasks, error recording and a transparent latency-versus-context curve.
Serving a very large multimodal MoE requires an inventory of full checkpoint storage, device count, tensor routing, KV-cache growth and acceptable concurrency. Use measured system load, not just parameter arithmetic. Operational and licensing risks need separate sign-off before any public paid service.
IMPLEMENTATION & EVALUATION
Evaluation checklist for reproducible and appropriately licensed Mistral Large 4 deployment. These are proposed tests, not observed OMW benchmark measurements.
Keep official checkpoint, revision SHA, tokenizer/config and weight manifest together.
Every retrieved weight shard is present, model and tokenizer configs match the pinned revision, and checksum provenance is logged.
Mixing a preview endpoint, NF4/NVFP4 conversion or older tokenizer creates a falsely attributed performance result.
Custom ecosystem terms include a separate agreement requirement for paid MaaS.
The deployment owner records the license text and a documented internal/legal determination for the actual proposed service.
Misclassifying custom MaaS restrictions as universal Apache/MIT commercial permission.
Published capacity and benchmark placement are not measured deployment behavior.
Run pinned text/image test prompts and long-context reference retrieval with accuracy, throughput, errors, device memory and concurrency recorded.
Conflating 52B active parameters with storage needs or promising 1M-token practical throughput without data.
Current benchmark placement changes over time and may use a hosted configuration.
Cite the independent leaderboard's exact evaluation date, index version, variant, reasoning setting and complete source URL.
Treating a publisher benchmark, preview API response or a stale leaderboard value as OMW's own checkpoint measurement.
Proposed tests, not OMW benchmark results. Read the official model card at the pinned source revision ↗ Inspect OMW field evidence ↗
PRIMARY & INDEPENDENT SOURCES
FAQ
Yes. The Mistral AI-owned Hugging Face repository exposes the checked original model's downloadable Safetensors artifacts and pinned source history. A separate quantized distribution is not the same exact checkpoint.
The Mistral Ecosystem Open-Weight License includes conditional commercial rights and expressly requires a separate Mistral agreement for paid Model-as-a-Service operations. Review the exact source license for your use case.
No. The 52B active-parameter description does not equal stored weights or peak memory; the released checkpoint is a very large MoE and Open Model Weights has not measured deployment memory.
The original weights are verified and the model is independently represented in Artificial Analysis open-weight intelligence comparisons. This editorial significance indicator is not an OMW benchmark score.
No. The checked config declares a token capacity, but useful retrieval fidelity, response quality, cost and latency at long lengths depend on model setup and need independent, task-specific tests. Open Model Weights has not performed those tests.
No. The October 6 API preview and October 29 downloadable checkpoint are distinct publication and delivery events. Compare only matching model variants, dates, inference settings, eval versions and published artifacts.
03 · PRACTICAL TECHNICAL PROFILE
Inspect weight-memory estimates, model lineage and runtime evidence separately from promotional or untested performance claims.
DEPLOYMENT PROFILE
These bars visualize weight-only decimal-GB estimates. They exclude KV cache, activations, runtime overhead, optimizer state and sharding.
LICENSE & USE
Commercial use under Mistral Ecosystem Open-Weight License conditions; operating a paid Model-as-a-Service requires a separate commercial agreement with Mistral.
This section summarizes the checked repository evidence. It is not legal advice. Inspect license evidence ↗
LINEAGE & PROVENANCE
Open Model Weights does not infer a hidden base model or training corpus when the checked sources do not declare one.
RUNTIME SIGNALS
Declared and model-card-mentioned runtimes are discovery signals. They are not treated as Open Model Weights runtime tests unless explicitly measured.
EVIDENCE TRAIL
This is where the article remains auditable: source links on one side, the registry's observed change history on the other.
Official Hugging Face repository SHA unchanged; prior field evidence retained.
Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.
04 · EVIDENCE AND UPDATES
Inspect dated source checks and any differences between published claims and the evidence; these checks are not independent model benchmarks.
SOURCE-BASED VISUAL BRIEF · LEADING
Three evidence-linked visual summaries: parameter scale, illustrative precision memory and verifiable provenance. No performance ranking is implied.
FRESHNESS & VERIFICATION
A verification check does not pretend the model changed. Open Model Weights updates the verification date after a real primary-source check, while factual-update dates move only when retained source evidence changes.
CONFLICT INTELLIGENCE
05 · SOURCE RECORD
Full repository-level attributes, licensing, format details and provenance are available below without hiding or rewriting the original data.
01 · WEIGHTS
consolidated-00001-of-00054.safetensorsconsolidated-00002-of-00054.safetensorsconsolidated-00003-of-00054.safetensorsconsolidated-00004-of-00054.safetensorsconsolidated-00005-of-00054.safetensorsconsolidated-00006-of-00054.safetensorsconsolidated-00007-of-00054.safetensorsconsolidated-00008-of-00054.safetensorsconsolidated-00009-of-00054.safetensorsconsolidated-00010-of-00054.safetensorsconsolidated-00011-of-00054.safetensorsconsolidated-00012-of-00054.safetensorsconsolidated-00013-of-00054.safetensorsconsolidated-00014-of-00054.safetensorsconsolidated-00015-of-00054.safetensorsconsolidated-00016-of-00054.safetensorsconsolidated-00017-of-00054.safetensorsconsolidated-00018-of-00054.safetensorsconsolidated-00019-of-00054.safetensorsconsolidated-00020-of-00054.safetensorsconsolidated-00021-of-00054.safetensorsconsolidated-00022-of-00054.safetensorsconsolidated-00023-of-00054.safetensorsconsolidated-00024-of-00054.safetensorsconsolidated-00025-of-00054.safetensorsconsolidated-00026-of-00054.safetensorsconsolidated-00027-of-00054.safetensorsconsolidated-00028-of-00054.safetensorsconsolidated-00029-of-00054.safetensorsconsolidated-00030-of-00054.safetensorsconsolidated-00031-of-00054.safetensorsconsolidated-00032-of-00054.safetensorsconsolidated-00033-of-00054.safetensorsconsolidated-00034-of-00054.safetensorsconsolidated-00035-of-00054.safetensorsconsolidated-00036-of-00054.safetensorsconsolidated-00037-of-00054.safetensorsconsolidated-00038-of-00054.safetensorsconsolidated-00039-of-00054.safetensorsconsolidated-00040-of-00054.safetensorsconsolidated-00041-of-00054.safetensorsconsolidated-00042-of-00054.safetensorsconsolidated-00043-of-00054.safetensorsconsolidated-00044-of-00054.safetensorsconsolidated-00045-of-00054.safetensorsconsolidated-00046-of-00054.safetensorsconsolidated-00047-of-00054.safetensorsconsolidated-00048-of-00054.safetensorsconsolidated-00049-of-00054.safetensorsconsolidated-00050-of-00054.safetensorsconsolidated-00051-of-00054.safetensorsconsolidated-00052-of-00054.safetensorsconsolidated-00053-of-00054.safetensorsconsolidated-00054-of-00054.safetensorsconsolidated.safetensors.index.json02 · LICENSE
Classification describes the checked license terms; it is not legal advice.
License / model-card evidence ↗03 · HARDWARE
Weight-only decimal-GB estimate from total parameter count; excludes KV cache, activations, runtime overhead, optimizer state and sharding.
04 · FORMATS
“Not observed” means no matching repository artifact/name/dtype signal was found; it does not prove no third-party conversion exists.
05 · LINEAGE
06 · TRAINING ASSETS
07 · RUNTIME SUPPORT
Open Model Weights separates model-repository evidence from external runtime-release evidence. An external support statement does not change the official model repository's weight formats and does not imply an Open Model Weights inference benchmark.
08 · SOURCES
09 · VERIFICATION HISTORY
Official Hugging Face repository SHA unchanged; prior field evidence retained.
Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.
MODELS FOR AI AGENTS · ROLE EVIDENCE
Multimodal reasoning model. A candidate for a reasoning or coordination role. No end-to-end agent performance is inferred from its model category.
The curated leading tier is not an agent-readiness rating. Model-specific tool-call success, long-horizon task accuracy and compatible runtime configurations are not independently established here.
Source-linked license: Mistral Ecosystem Open-Weight License v26.10 (allowed-with-conditions commercial-use classification; see original terms).
Context: 1048576 configured tokens.
BF16 weights: 2100.0 estimated decimal GB, not runtime VRAM.
Compare agent model candidates → · Agent evaluation method → · Inspect original model repository ↗
These records are separately sourced. They do not replace independent model-plus-runtime tests.
No model-specific agent capability evaluation is recorded for this checkpoint. Missing evidence is unknown, not a failed task.
Read evidence details and benchmarks → · Use the agent model selector →