Mistral Large 3 2512 vs Qwen3 Max
Mistral AIQwenboth via one key, provider price + 3%
Both are large, non-reasoning-by-default text models released within three months of each other, sharing the same 262,144-token context window. Mistral Large 3 2512 is Mistral's flagship sparse MoE (41B active of 675B total params, Apache 2.0), while Qwen3 Max is Qwen's flagship dense-ish successor with a June 2025 knowledge cutoff. The split shows up mostly in Design Arena rankings and price, not context or modality — Mistral adds image and file input, Qwen3 Max is text-only.
Spec vs spec
| Spec | Mistral Large 3 2512 | Qwen3 Max |
|---|---|---|
| Context window | 262K | 262K |
| Max output | — | 33K |
| Input modalities | text, image, file | text |
| Output modalities | text | text |
| Knowledge cutoff | — | Jun 30, 2025 |
| Released | Dec 1, 2025 | Sep 23, 2025 |
| Reasoning | — | optional |
Pricing
Per 1M tokens. Provider price plus the flat 3% fee — the sum is what you pay.
mistralai/mistral-large-2512
Input · 1M tokens
$0.500 + 3%$0.515
Output · 1M tokens
$1.50 + 3%$1.54
Cache read · 1M tokens
$0.050 + 3%$0.052
FEE — FLAT, EVERY MODEL3%
qwen/qwen3-max
Input · 1M tokens
$0.780 + 3%$0.803
Output · 1M tokens
$3.90 + 3%$4.02
Cache read · 1M tokens
$0.156 + 3%$0.161
Cache write · 1M tokens
$0.975 + 3%$1.00
FEE — FLAT, EVERY MODEL3%
One workload, priced on both
10M input + 2M output tokens at each model's price, flat 3% fee included.
mistralai/mistral-large-2512Cheaper
$8.24
$8.00 provider + 3%
qwen/qwen3-max
$16.07
$15.60 provider + 3%
Benchmarks
Design Arena categories where both models have results. Higher Elo and lower rank win.
| Mistral Large 3 2512 | Qwen3 Max | |||
|---|---|---|---|---|
| Category | Elo | Rank | Elo | Rank |
| 3D | 1176 | #51 | 1150 | #62 |
| ASCII art | 1115 | #43 | 1175 | #32 |
| Code | 1191 | #56 | 1159 | #66 |
| Data viz | 1180 | #55 | 1149 | #64 |
| Game dev | 1146 | #65 | 1160 | #62 |
| SVG | 1050 | #62 | 1069 | #60 |
| UI components | 1157 | #59 | 1132 | #67 |
| Websites | 1205 | #53 | 1161 | #66 |
Head-to-head preference voting. How we filter and rank
Design Arena: who wins which category
Mistral Large 3 2512 takes 5 of 8 Design Arena categories: 3d (elo 1176 vs 1150), codecategories (1191 vs 1159), dataviz (1180 vs 1149), uicomponent (1157 vs 1132), and website (1205 vs 1161). Qwen3 Max wins the remaining 3: asciiart (1175 vs 1115), gamedev (1160 vs 1146), and svg (1069 vs 1050).
For anything touching data visualization, UI components, or full website generation, Mistral's elo lead is consistent and not close — it's a 20-45 point gap in every category it wins. Qwen3 Max's wins are narrower except asciiart, where it beats Mistral by 60 elo points. If your workload is mostly ASCII rendering or game-dev scaffolding, that's the one place Qwen3 Max has a real edge.
Pricing math on a real workload
Provider list price: Mistral Large 3 2512 runs $0.50/M input and $1.50/M output. On OpenKey that's $0.50 × 1.03 = $0.515/M input and $1.50 × 1.03 = $1.545/M output. Qwen3 Max runs $0.78/M input and $3.90/M output, or $0.78 × 1.03 = $0.8034/M input and $3.90 × 1.03 = $4.017/M output on OpenKey.
On a 10M-input / 2M-output workload, Mistral Large 3 2512 costs $8.00 total versus Qwen3 Max's $15.60 — almost double. Mistral's input tokens are priced at a 0.64 ratio to Qwen3 Max's (roughly 36% cheaper per input token), and the output-token gap is even wider. If you're running high-volume batch jobs, that difference compounds fast.
Context and modality
Both models share an identical 262,144-token context window — the context ratio between them is 1.0, so neither has a long-document advantage. Where they diverge is input modality: Mistral Large 3 2512 accepts text, image, and file input; Qwen3 Max is text-only. If your pipeline needs to feed in screenshots, PDFs, or scanned documents directly, Mistral is the only one of the two that can take them without a separate OCR or vision step. Qwen3 Max caps output at 32,768 tokens; Mistral has no stated max-completion limit in this record.
Tool use and structured output
Both support `tool_choice`, `tools`, `structured_outputs`, `response_format`, `temperature`, `top_p`, `seed`, and `presence_penalty` — parity for anyone building agent loops or JSON-mode pipelines. Mistral additionally supports `frequency_penalty` and `stop` sequences. Qwen3 Max additionally supports `logprobs` and `top_logprobs`, which matters if you need token-level confidence scores for classification or ranking tasks. Neither model is marked as mandatory-reasoning; Qwen3 Max has an optional reasoning flag, Mistral has none listed.
Which model for which job
| Use case | Pick | Why |
|---|---|---|
| UI component generation | Mistral Large 3 2512 | 1157 elo vs Qwen3 Max's 1132 on Design Arena uicomponent |
| Full website generation | Mistral Large 3 2512 | 1205 elo vs Qwen3 Max's 1161, the largest gap outside asciiart |
| High-volume batch processing | Mistral Large 3 2512 | $8.00 vs $15.60 on a 10M-in/2M-out workload |
| ASCII art / text-based rendering | Qwen3 Max | 1175 elo vs Mistral's 1115, the biggest category lead either model has |
| Game dev prototyping | Qwen3 Max | 1160 elo vs Mistral's 1146 on Design Arena gamedev |
| Document/image-in-prompt workflows | Mistral Large 3 2512 | supports image and file input; Qwen3 Max is text-only |
Questions
- Which model is cheaper to run at scale?
- Mistral Large 3 2512, by a wide margin. On OpenKey it's $0.515/M input and $1.545/M output versus Qwen3 Max's $0.8034/M input and $4.017/M output. A 10M-input/2M-output job costs $8.00 on Mistral versus $15.60 on Qwen3 Max.
- Does either model support image input?
- Mistral Large 3 2512 does — it takes text, image, and file input. Qwen3 Max is text-only, both in and out. If your app needs to read screenshots or PDFs directly, Mistral is the only option of the two.
- Which model wins more Design Arena categories?
- Mistral Large 3 2512 wins 5 of 8: 3d, codecategories, dataviz, uicomponent, and website. Qwen3 Max wins the other 3: asciiart, gamedev, and svg, with asciiart being its biggest lead (1175 vs 1115 elo).
- Can I use both models through one API key?
- Yes — both are available on OpenKey with a single key, and pricing is provider list price plus a flat 3% fee (already reflected in the $0.515/$1.545 and $0.8034/$4.017 per-million-token figures above).