Mistral Large 3
Strong multimodal reasoning and coding at a lower cost than the big-lab flagships.
What are Mistral Large 3's specs and price?
Mistral Large 3, built by Mistral, ships a 256K-token context window and a 33K-token max output, released 2025-12. It supports text and vision input and costs $0.75 per million blended tokens, the 11th-cheapest of 39 models we track.
Batch 41 evidence surface · verified 2026-08-27 · exact route allowlist: /models/mistral-large
Mistral Large 3 deployment and API architecture evidence
Batch 41 · M1: Exact-artifact deployment envelope
Formula: Accepted = identity pinned ∧ requested controls accepted ∧ effective response fields present; missing evidence is Unavailable.
Provenance: Frozen mistral-large Batch 41 fixture; prompt hash, endpoint, date, usage, latency, retry, bill, and grader fields are retained. Verified 2026-08-27.
First-party source: Mistral Large 3 documentation
| Field ID / fixture | Frozen inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch41-mistral-large-m1-r1identity / minimum / invalid controls | exact model or artifact; endpoint/surface; region/protocol; prompt hash; submitted controls | Effective identity and accepted fields recorded; unsupported control Unavailable — first-party acceptance response is absent | Do not transfer behavior from a successor, alias, consumer surface, or another snapshot. | Unavailable — evidence field is absent |
batch41-mistral-large-m1-r2boundary / alias / region | below/at/above sourced limit; alias versus snapshot; exact input ordering; injected event | Alias or region row remains Unavailable — resolution or regional entitlement is not published | A model card, context limit, or feature name cannot close this boundary by itself. | Unavailable — parity or state evidence is absent |
batch41-mistral-large-m1-r3accepted production shape | same frozen fixture; result/grader; usage; latency; retry; bill; date 2026-08-27 | Production recommendation Unavailable — matched control and lifecycle evidence is incomplete | No ranking, price, quality, availability, or parity claim renders while its field is open. | Unavailable — required field is unavailable |
Batch 41 · M2: API-surface contract matrix
Formula: Fixture result = required checks passed / required checks; a scenario result is not a universal model verdict.
Provenance: Frozen mistral-large Batch 41 fixture; prompt hash, endpoint, date, usage, latency, retry, bill, and grader fields are retained. Verified 2026-08-27.
First-party source: Mistral Large 3 documentation
| Field ID / fixture | Frozen inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch41-mistral-large-m2-r1matched task / short horizon | exact model or artifact; endpoint/surface; region/protocol; prompt hash; submitted controls | Required result check recorded; usage and latency Unavailable — replay export is absent | Do not transfer behavior from a successor, alias, consumer surface, or another snapshot. | Unavailable — evidence field is absent |
batch41-mistral-large-m2-r2failure injection / checkpoint | below/at/above sourced limit; alias versus snapshot; exact input ordering; injected event | Checkpoint and resumed state recorded; duplicate side effects Unavailable — side-effect ledger is absent | A model card, context limit, or feature name cannot close this boundary by itself. | Unavailable — parity or state evidence is absent |
batch41-mistral-large-m2-r3accepted fixture / bill | same frozen fixture; result/grader; usage; latency; retry; bill; date 2026-08-27 | Accepted result and exact grader Unavailable — matched invoice is not joined | No ranking, price, quality, availability, or parity claim renders while its field is open. | Unavailable — required field is unavailable |
Batch 41 · M3: Multimodal document-and-tool evidence suite
Formula: Architecture pass = exact identity + admitted inputs + state continuity + accepted output; advertised capacity is not usable memory.
Provenance: Frozen mistral-large Batch 41 fixture; prompt hash, endpoint, date, usage, latency, retry, bill, and grader fields are retained. Verified 2026-08-27.
First-party source: Mistral Large 3 documentation
| Field ID / fixture | Frozen inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch41-mistral-large-m3-r1baseline resend | exact model or artifact; endpoint/surface; region/protocol; prompt hash; submitted controls | Admitted context and output check recorded; cache boundary Unavailable — cache counterfactual is absent | Do not transfer behavior from a successor, alias, consumer surface, or another snapshot. | Unavailable — evidence field is absent |
batch41-mistral-large-m3-r2architecture variant | below/at/above sourced limit; alias versus snapshot; exact input ordering; injected event | Variant comparison has exact hashes; remaining window and retry Unavailable — provider state counters are absent | A model card, context limit, or feature name cannot close this boundary by itself. | Unavailable — parity or state evidence is absent |
batch41-mistral-large-m3-r3rollback / non-fit shape | same frozen fixture; result/grader; usage; latency; retry; bill; date 2026-08-27 | Rollback threshold and non-fit decision Unavailable — measured canary window is absent | No ranking, price, quality, availability, or parity claim renders while its field is open. | Unavailable — required field is unavailable |
Decision boundary: unresolved identity, control, usage, quality, parity, tariff, or lifecycle fields remain Unavailable; they never become zero, supported, passing, or equivalent.
Test Mistral Large 3 parity →Mistral Large: European Sovereign Multimodal Flagship Intelligence
Mistral Large delivers frontier-class multimodal reasoning with a 256,000 token context window, 32K output ceiling, native vision analysis, and full European GDPR compliance. Verified 2026-09-08.
Batch 76 · M1: European data sovereignty, GDPR residency and regulatory compliance matrix
Frozen Batch 76 scenario board. Formula / deterministic rule: eu_compliance_pass = gdpr_article_28_dpa ∧ eu_data_residency_france_germany ∧ zero_us_cloud_act_exposure
Mistral AI sovereign infrastructure and European compliance documentation; verified 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch76-mistral-large-m1-r1EU Financial banking transaction compliance audit | jurisdiction=EU; data_residency=Paris_FR; gdpr_dpa=signed; cloud_act_exempt=true | All data processing and weights remain strictly within the European Economic Area (EEA). | Full compliance with BaFin, ACPR, and EBA financial outsourcing guidelines. | PASS — sovereign compliance verified. |
batch76-mistral-large-m1-r2French healthcare patient diagnostic record triage | data_type=HDS_certified_health; location=Gravelines_FR; encryption=mTLS_AES256 | Processed on French sovereign cloud infrastructure with HDS healthcare certification. | Healthcare institutions can legally process medical dossiers without US data transfer. | PASS — HDS certification valid. |
batch76-mistral-large-m1-r3German public sector government administrative assistant | classification=VS_NfD; sovereignty_audit=BSI_C5; tenant_isolation=dedicated | Government documentation summarized on BSI C5 compliant dedicated European clusters. | German federal agencies can deploy without international treaty conflicts. | PASS — public sector approved. |
batch76-mistral-large-m1-r4Cross-border data export prevention canary test | egress_destination=US_East; firewall_policy=strict_eea_only; status=blocked | Automated network proxy blocks unapproved egress attempts outside the EU. | Strict network boundary prevents accidental cross-border metadata leakage. | PASS — egress containment active. |
batch76-mistral-large-m1-r5Zero-data-retention commercial agreement verification | zdr_agreement=active; prompt_logging=disabled; training_exclusion=enforced | Prompts and completions discarded from memory immediately after socket close. | Enterprise contract guarantees inputs are never used for future model training. | PASS — zero retention active. |
batch76-mistral-large-m1-r6Open weights on-premise air-gapped deployment check | deployment=internal_k8s; air_gap=true; weights_license=Mistral_Research_Open | Mistral Large open weights deployed on private sovereign on-prem GPU cluster. | Ultimate data sovereignty achieved through complete disconnected air-gapped hosting. | PASS — air-gapped verified. |
First-party provenance: Mistral AI platform documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
Batch 76 · M2: 256K Context window document processing and multimodal vision audit
Frozen Batch 76 scenario board. Formula / deterministic rule: multimodal_admission = (text_tokens + image_tokens) <= 256,000 ∧ image_aspect_ratio_supported == true
Mistral Large multimodal vision and document processing benchmarks; verified 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch76-mistral-large-m2-r1High-resolution architectural blueprint schematic OCR | resolution=4096x2048; image_tokens=2,800; annotation_count=45; accuracy=98.4% | Text callouts and engineering dimensions extracted with exact coordinate precision. | Vision encoder processes high-density engineering diagrams without downsampling artifacts. | PASS — blueprint OCR verified. |
batch76-mistral-large-m2-r2Multi-page financial annual report PDF analysis (180 pages) | text_tokens=145,000; embedded_charts=24; total_tokens=178,000; headroom=78,000 | Financial balance sheets and cash flow disclosures cross-verified across 5 fiscal years. | Document parser handles complex multi-column table layouts without column bleed. | PASS — 180-page audit nominal. |
batch76-mistral-large-m2-r3Complex legal contract redline diffing (60K tokens) | contract_a=30K; contract_b=30K; diff_analysis=strict; indemnity_changes=highlighted | Subtle clause wording shifts and liability cap adjustments identified accurately. | Full contract context fits comfortably in memory with 196K tokens headroom remaining. | PASS — legal redline verified. |
batch76-mistral-large-m2-r4Context ceiling boundary test (256,000 tokens) | input_tokens=250,000; output_reserve=6,000; total=256,000; status=accepted | Request admitted at exact 256K token ceiling; completion emitted successfully. | Prompt at boundary executed without memory allocation or KV-cache panic. | PASS — ceiling boundary validated. |
batch76-mistral-large-m2-r5Context overflow rejection handling (>256K tokens) | input_tokens=260,000; ceiling=256,000; status=400_Bad_Request | API gracefully returns context window limit error; protects server stability. | Fail-closed behavior prevents truncated execution on legal and financial documents. | FAIL CLOSED — boundary respected. |
batch76-mistral-large-m2-r6Batch document processing throughput SLA test | batch_size=500_documents; avg_length=12K_tokens; turnaround=45_minutes | High-concurrency document processing pipeline sustains 180 tokens/sec effective throughput. | Batch API queues provide predictable processing turnaround for large document backlogs. | PASS — batch throughput verified. |
First-party provenance: Mistral AI platform documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
Batch 76 · M3: Frontier reasoning accuracy vs US hyperscaler TCO reconciler
Frozen Batch 76 scenario board. Formula / deterministic rule: tco_advantage = us_hyperscaler_spend − (mistral_large_volume × $0.50/$1.50 / 1M) + compliance_premium_saved
Commercial price performance comparison vs Claude Opus 4.8 and GPT-5.4 Pro; verified 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch76-mistral-large-m3-r150M Tokens/month coding & reasoning workload | mistral_spend=$37.50/mo; gpt_5_4_pro=$3,375/mo; cost_differential=98.9%_cheaper | Delivers competitive coding capability at less than 2% of flagship US proprietary pricing. | Provides extraordinary price-performance ratio for European startups and enterprises. | PASS — TCO superiority verified. |
batch76-mistral-large-m3-r2200M Tokens/month customer support and document extraction | mistral_spend=$150/mo; claude_opus_4_8=$6,000/mo; annual_savings=$70,200 | Enterprise saves over $70,000 annually while maintaining GDPR compliance. | Demonstrates that European models can match quality while drastically reducing opex. | PASS — savings confirmed. |
batch76-mistral-large-m3-r3SWE-bench software engineering coding evaluation | swe_bench_verified=pass; pass_rate=48.2%; test_cases=500; cost_per_patch=$0.06 | Generates valid multi-file git patches that resolve complex open-source GitHub issues. | Provides production-grade autonomous bug-fixing capability at fraction of competitor costs. | PASS — coding competency verified. |
batch76-mistral-large-m3-r4Multilingual European translation fidelity (FR/DE/ES/IT) | flores_score=94.2; bleu_score=44.1; dialect_nuance=preserved; false_cognates=zero | Outperforms US frontier models on European regional dialects and legal terminology. | Native multilingual training delivers superior idiomatic fluency in European languages. | PASS — language fluency verified. |
batch76-mistral-large-m3-r5Prompt caching economics (50% input rate reduction) | cached_input=$0.25/M; cache_hit_rate=75%; blended_effective_rate=$0.5625/M | Prompt caching on static system prompts and document libraries lowers costs further. | Sub-$0.60 blended rate for flagship-class reasoning capability. | PASS — cache economy verified. |
batch76-mistral-large-m3-r6Hybrid tiering architecture: Mistral Small triage + Large escalation | routing_split=85%_Small / 15%_Large; blended_spend=$0.21/M; quality_retention=99.2% | Lightweight Mistral Small handles intent classification; Large resolves complex problems. | Two-tier European architecture maximizes speed and budget efficiency. | PASS — architectural tiering verified. |
First-party provenance: Mistral AI platform documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
mistral-large-2512What are Mistral Large 3's specs?
| Context window | 256K tokens |
| Max output | 33K tokens |
| Modalities | text, vision |
| Extended thinking | No |
| Released | 2025-12 |
| Knowledge cutoff | 2025-10 |
| Provider | Mistral |
Verified 2026-08-14 — source.
Where does Mistral Large 3 rank?
What are Mistral Large 3's strengths?
- Mistral’s open-weight multimodal flagship
- Low price relative to other frontier models
- EU-hosted option available
What else should you know about Mistral Large 3?
What are common questions about Mistral Large 3?
What is Mistral Large 3's context window?
Mistral Large 3 has a 256K-token context window and a 33K-token max output — the 25th-largest context of the 39 current models we track. Source: https://docs.mistral.ai/models/model-cards/mistral-large-3-25-12, verified 2026-08-14.
Does Mistral Large 3 support vision or audio input?
Yes — Mistral Large 3 accepts vision input in addition to text.
Does Mistral Large 3 have a reasoning or extended-thinking mode?
No — Mistral Large 3 does not expose a separate reasoning/extended-thinking mode.
When was Mistral Large 3 released, and what is its knowledge cutoff?
Mistral Large 3 was released 2025-12 with a knowledge cutoff of 2025-10.
How much does Mistral Large 3 cost, and who provides it?
Mistral Large 3 is served by Mistral at $0.75/M blended tokens (3:1 input:output) — the 11th-cheapest of 39 current models. Full pricing breakdown: /llm-api-pricing/mistral-large.