Today's price · per 1M tokens
Input
$0.700
per 1M tokens
Output
$2.10
per 1M tokens
Blended
$1.05
blended $/1M — 3:1 weighted input:output
Cached input
$0.070
10% of input — prompt caching
How this price is scoped: Mistral cut the price on 29 October 2026: its API price list and the model page now charge $0.70 input / $0.07 cached input / $2.10 output per 1M tokens, with no struck-through original beside it. The launch rate was $1.36 / $0.14 / $4.18.
Price receipt
We read Mistral's own pricing page on and found Mistral Large 4 listed with a price on the same row — that read is where the number above comes from. We keep the page text we read; its content hash is b146735b1c.
Overview
Mistral's new flagship, launched 2026-10-06 as a public preview open to anyone on Mistral Studio — a 1.05T-parameter MoE (52B active) with image input and a 1M-token context, unifying instruction, reasoning and agentic use in one model. Mistral has published the weights on Hugging Face under its Mistral Ecosystem Open Weight License. $0.70/$2.10 per 1M tokens.
Capabilities
struck through = not supportedBenchmark performance
accuracy % · higher is betterSources: Epoch AI (epoch.ai/benchmarks, CC BY 4.0) (GPQA Diamond)
Source: Epoch AI (epoch.ai/benchmarks), CC BY 4.0
- AA Intelligence Index: 26.1 (high effort) (Artificial Analysis composite across reasoning, knowledge and coding evals)
- AA-Omniscience Index: 5.1 (high effort) (−100–100)
- GDPval-AA v2: 1433.8 ELO (Real-world work tasks, human baseline = 1000)
- Arena (arena.ai): 1422.7 ELO (Human preference)
- Arena WebDev: 1573.2 ELO (Human preference on building web apps)
- Arena Vision: 1233.2 ELO (Human preference on image-input prompts)
- Arena Agent: -0.071 (Score from real agent sessions (IPS); higher is better)
- Arena Agent cost per task: $0.26 (Median measured cost of one real Agent Arena task)
- Epoch Capabilities Index: 152.7 (CI 150.8–154.7) (Epoch AI's index fitted across many benchmarks)
- FrontierMath Tiers 1–3: 58.6% (high effort) (Research-level maths (funded by OpenAI, run by Epoch); best run across the effort settings Epoch tested)
- FrontierMath Tier 4: 26.8% (high effort) (Hardest research maths; best run across the effort settings Epoch tested)
Source: Artificial Analysis (artificialanalysis.ai)
- HLE-Diamond: 17.3% (high effort, measured 2026-11-22) — A revised 1,000-question set of Humanity's Last Exam, half reasoning, half knowledge
- CritPt v2, hard subset: 27% (high effort, measured 2026-11-22) — The hardest research-level physics problems in CritPt v2, solved with a code sandbox
- FrontierCode-AA 2.0: 15% (high effort, measured 2026-11-22) — Coding tasks in real open-source repositories, graded on whether a maintainer would merge the change
- Terminal-Bench-Science: 4.3% (high effort, measured 2026-11-22) — Expert-written science research workflows completed in a terminal
- GDP.pdf: 17.6% (high effort, measured 2026-11-22) — Tasks over long professional documents; share of tasks that pass every criterion
- AutomationBench-AA: 57.8% (high effort, measured 2026-11-22) — Multi-step work across simulated business apps, without breaking the rules set
- MLCR-AA: 17.8% (high effort, measured 2026-11-22) — Reasoning over long, fragmented medical records; pass rate
- Terminal-Bench 4.0: 24.7% (high effort, measured 2026-11-22)
- Cost to run the AA Intelligence Index: $2.06 per task, 20.5 min per task (high effort)
- Measured by AA: 66.6 tokens/sec output, 2.95s to first token, 64.72s to first answer token (high effort)
- Ranks #85 of 207 comparably-measured models by percentile score, across 5 independent measurements
- Ranks #55 of 207 comparably-tested models by normalized performance per dollar
- Strongest at GPQA Diamond — 88.1%, #61 of 128
66.6 tokens/sec output 2.95s latency to first token (TTFT)
Throughput: Artificial Analysis, measured 2026-11-22 · Source: Artificial Analysis (artificialanalysis.ai)
Source: · updated · See full rankings →
Specifications
- Provider
- Mistral
- Context window
- 1M tokens
- Modality
- text, image
- Parameters
- 1.05T
- Open source
- No — proprietary
- Released
- Status
- Preview
- Last updated
- Tags
Availability verified: — listed on Mistral's own page