DeepSeek V4 Pro vs Mistral Large 3 675B Instruct
DeepSeek V4 Pro (2026) and Mistral Large 3 675B Instruct (2025) are frontier reasoning models from DeepSeek and MistralAI. DeepSeek V4 Pro ships a 1m-token context window, while Mistral Large 3 675B Instruct ships a 128k-token context window. On MMLU PRO, DeepSeek V4 Pro leads by 2 pts. This comparison covers specs, pricing, API access, capabilities, benchmarks, input and output token costs, and production fit for coding and agent workloads.
DeepSeek V4 Pro fits 8x more tokens; pick it for long-context work and Mistral Large 3 675B Instruct for tighter calls.
Decision scorecard
Local evidence first| Signal | DeepSeek V4 Pro | Mistral Large 3 675B Instruct |
|---|---|---|
| Best for | reasoning-heavy apps, tool-calling agents, and long-context analysis | multimodal apps and provider-routed production |
| Decision fit | Coding, RAG, and Agents | Coding, RAG, and Agents |
| Context window | 1m | 128k |
| Cheapest output | $0.87/1M tokens | $1.50/1M tokens |
| Provider routes | 5 tracked | 5 tracked |
| Shared benchmarks | MMLU PRO leader | 4 rows |
Decision tradeoffs
- DeepSeek V4 Pro holds a shared-benchmark lead on MMLU PRO, ahead by 2 points.
- DeepSeek V4 Pro has the larger context window for long prompts, retrieval packs, or transcript analysis.
- DeepSeek V4 Pro has the lower cheapest tracked output price at $0.87/1M tokens.
- DeepSeek V4 Pro uniquely exposes Reasoning, Function calling, and Tool use in local model data.
- Local decision data tags DeepSeek V4 Pro for Coding, RAG, and Agents.
- Mistral Large 3 675B Instruct holds a shared-benchmark lead on HumanEval, ahead by 15.2 points.
- Mistral Large 3 675B Instruct uniquely exposes Vision and Multimodal in local model data.
- Local decision data tags Mistral Large 3 675B Instruct for Coding, RAG, and Agents.
Monthly cost at traffic
Estimate token spend from the cheapest tracked input and output route or tier on this page.
DeepSeek V4 Pro
$566
Cheapest tracked route/tier: DeepSeek Platform
Mistral Large 3 675B Instruct
$775
Cheapest tracked route/tier: AWS Bedrock
Estimated monthly gap: $210. Batch, cache, alternate speed tiers, and negotiated pricing are excluded from this local estimate.
Switch friction
- Provider overlap exists on Vercel AI Gateway; start route-level A/B tests there.
- Mistral Large 3 675B Instruct is $0.63/1M tokens higher on cheapest tracked output pricing, so quality gains need to justify the spend.
- Check replacement coverage for Reasoning, Function calling, and Tool use before moving production traffic.
- Mistral Large 3 675B Instruct adds Vision and Multimodal in local capability data.
- Provider overlap exists on Vercel AI Gateway; start route-level A/B tests there.
- DeepSeek V4 Pro is $0.63/1M tokens lower on cheapest tracked output pricing before cache, batch, or negotiated discounts.
- Check replacement coverage for Vision and Multimodal before moving production traffic.
- DeepSeek V4 Pro adds Reasoning, Function calling, and Tool use in local capability data.
Specs
| Specification | ||
|---|---|---|
| Released | 2026-04-24 | 2025-12-01 |
| Context window | 1m | 128k |
| Parameters | 1.6T | 675B |
| Architecture | Mixture of Experts (MoE) with CSA+HCA hybrid attention | decoder only |
| License | MIT(OSI) | Mistral License |
| Openness | Open source | Open weights |
| Commercial use | Commercial use allowed | Non-commercial only |
| Knowledge cutoff | - | 2024-11 |
Pricing and availability
| Pricing attribute | DeepSeek V4 Pro | Mistral Large 3 675B Instruct |
|---|---|---|
| Input price | $0.43/1M tokens | $0.50/1M tokens |
| Output price | $0.87/1M tokens | $1.50/1M tokens |
| Providers |
Capabilities
| Capability | DeepSeek V4 Pro | Mistral Large 3 675B Instruct |
|---|---|---|
| Vision | No | Yes |
| Multimodal | No | Yes |
| Reasoning | Yes | No |
| Function calling | Yes | No |
| Tool use | Yes | No |
| Structured outputs | Yes | Yes |
| Code execution | No | No |
| IDE integration | No | No |
| Computer use | No | No |
| Parallel agents | No | No |
Benchmarks
| Benchmark | DeepSeek V4 Pro | Mistral Large 3 675B Instruct |
|---|---|---|
| MMLU PRO | 87.5 | 85.5 |
| Google-Proof Q&A | 90.1 | 43.9 |
| LiveCodeBench | 93.5 | 82.8 |
| HumanEval | 76.8 | 92.0 |
Deep dive
On shared benchmark coverage, MMLU PRO has DeepSeek V4 Pro at 87.5 and Mistral Large 3 675B Instruct at 85.5, with DeepSeek V4 Pro ahead by 2 points; Google-Proof Q&A has DeepSeek V4 Pro at 90.1 and Mistral Large 3 675B Instruct at 43.9, with DeepSeek V4 Pro ahead by 46.2 points; LiveCodeBench has DeepSeek V4 Pro at 93.5 and Mistral Large 3 675B Instruct at 82.8, with DeepSeek V4 Pro ahead by 10.7 points. The largest visible gap is 46.2 points on Google-Proof Q&A, which matters most when that benchmark mirrors your workload. Treat isolated benchmark wins as directional, because provider routing, prompt style, and tool access can move real application results.
The capability footprint differs most on vision: Mistral Large 3 675B Instruct, multimodal input: Mistral Large 3 675B Instruct, reasoning mode: DeepSeek V4 Pro, function calling: DeepSeek V4 Pro, and tool use: DeepSeek V4 Pro. Both models share structured outputs, so the practical split is not just feature count. Use those differences to decide whether the page is about raw model quality, agentic coding support, multimodal ingestion, or predictable structured API behavior.
For cost, DeepSeek V4 Pro lists $0.43/1M input and $0.87/1M output tokens on the cheapest tracked provider, while Mistral Large 3 675B Instruct lists $0.50/1M input and $1.50/1M output tokens on the cheapest tracked provider. A 70/30 input-output blend puts DeepSeek V4 Pro lower by about $0.23 per million blended tokens. Availability is 5 providers versus 5, so concentration risk also matters.
Choose DeepSeek V4 Pro when reasoning depth, larger context windows, and lower input-token cost are central to the workload. Choose Mistral Large 3 675B Instruct when vision-heavy evaluation are more important. For production, rerun your own prompts through the exact provider, region, and tool stack you plan to ship.
FAQ
Which has a larger context window, DeepSeek V4 Pro or Mistral Large 3 675B Instruct?
DeepSeek V4 Pro supports 1m tokens, while Mistral Large 3 675B Instruct supports 128k tokens. That gap matters most for long documents, large codebases, retrieval-heavy agents, and conversations where earlier context must remain visible.
Which is cheaper, DeepSeek V4 Pro or Mistral Large 3 675B Instruct?
DeepSeek V4 Pro is cheaper on tracked token pricing. DeepSeek V4 Pro costs $0.43/1M input and $0.87/1M output tokens. Mistral Large 3 675B Instruct costs $0.50/1M input and $1.50/1M output tokens. Provider discounts or batch pricing can still change the final bill.
Is DeepSeek V4 Pro or Mistral Large 3 675B Instruct open source?
DeepSeek V4 Pro is listed under MIT. Mistral Large 3 675B Instruct is listed under Mistral License. License labels affect whether you can self-host, redistribute weights, or rely only on hosted APIs, so confirm the upstream license before deployment.
Which is better for vision, DeepSeek V4 Pro or Mistral Large 3 675B Instruct?
Mistral Large 3 675B Instruct has the clearer documented vision signal in this comparison. If vision is mission-critical, validate it against the provider endpoint because model-level support and API-level exposure can differ.
Which is better for multimodal input, DeepSeek V4 Pro or Mistral Large 3 675B Instruct?
Mistral Large 3 675B Instruct has the clearer documented multimodal input signal in this comparison. If multimodal input is mission-critical, validate it against the provider endpoint because model-level support and API-level exposure can differ.
Where can I run DeepSeek V4 Pro and Mistral Large 3 675B Instruct?
DeepSeek V4 Pro is available on DeepSeek Platform, Fireworks AI, OpenRouter, Vercel AI Gateway, and Novita AI. Mistral Large 3 675B Instruct is available on AWS Bedrock, NVIDIA NIM, Mistral AI Studio, Microsoft Foundry, and Vercel AI Gateway. Provider coverage can affect latency, region availability, compliance posture, and fallback options.
Continue comparing
Last reviewed: 2026-06-04. Data sourced from public model cards and provider documentation.