Open MoE flagship with million-token context for coding and long agent runs
DeepSeek V4 Pro is offered by 77 providers on this page. Public prices are shown for 68 of them. Its official list price is $0.66 per 1M input tokens and $1.98 per 1M output tokens. The lowest paid channel is routing.run at $0.348 / $0.696 per 1M, about 6.9× below the list price. That channel is a third-party gateway, so confirm its availability and rate limits before depending on it. It also has 3 free ($0) channels and 4 covered by a paid subscription; free tiers usually carry rate limits, and subscription-covered access bills $0 per token only after the subscription fee.
The context window is 1,000,000 tokens at the reference host, but hosts report different limits, from 128,000 to 1,050,000, so the usable window depends on the provider you pick. It supports reasoning, tool use, and structured output. The weights are open, so it can also be self-hosted or served through a gateway of your choice. Providers report the capability flags inconsistently, so verify a specific feature against the host you plan to use. Its training knowledge cuts off at 2025-05.
| Provider | Tier | Input | Output | Cache read | Cache write | Context | Output limit | Status |
|---|---|---|---|---|---|---|---|---|
| SenseNova (China) | Gateway | Free | $0 | — | 1,048,576 ⚠ | 65,536 | ||
| Alibaba Token Plan | Gateway | Subscription | $0 | $0 | 1,000,000 | 384,000 | ||
| Alibaba Token Plan (China) | Gateway | Subscription | $0 | $0 | 1,000,000 | 384,000 | ||
| UnoRouter deepseek-v4-pro:free | Gateway | Free | — | — | 1,000,000 | 384,000 | ||
| Volcengine Ark Coding Plan | Gateway | Subscription | $0 | — | 1,000,000 | 384,000 | ||
| SCNet Token Plan DeepSeek-V4-Pro | Gateway | Subscription | $0 | — | 1,000,000 | 384,000 | ||
| Kenari | Gateway | Free | — | — | 1,000,000 | 384,000 | ||
| routing.run | Gateway | $0.35 | $0.70 | — | — | 1,000,000 | 64,000 | |
| CrofAI | Gateway | $0.35 | $0.80 | $0.003 | — | 1,000,000 | 131,072 | |
| EBCloud DeepSeek-V4-Pro | Gateway | $0.43 | $0.86 | — | — | 1,000,000 | 384,000 | |
| DeepSeekOfficial | First-party | $0.66 | $1.98 | $0.022 | — | 1,000,000 | 393,216 | |
Sorted by blended price (input×0.75 + output×0.25) asc. The official channel always shows regardless of rank. Whether a gateway's low price is actually usable can't be verified.
Interleaved thinking (reasoning between tool calls) is declared by 52 of 77 providers.
11 / 77 providers expose no reasoning control (reasoning_options: []).
7 more channels offer $0 (SenseNova (China), Alibaba Token Plan, Alibaba Token Plan (China) etc.); free tiers usually have rate limits and no SLA, excluded from ranking.
| Name | Conditions | Score | Metric | Source |
|---|---|---|---|---|
| SWE-Bench Verified | — | 80.6 | resolved | Source ↗ |
| Artificial Analysis Coding Agent Index | harness: Claude Code · variant: high | 50.1 | average pass@1 | Source ↗ |
| SWE-Atlas Codebase QnA | harness: Claude Code · variant: high | 67.8 | pass@1 | Source ↗ |
| SWE-Bench Pro | harness: Claude Code · variant: high · dataset: hard-aa | 18 | pass@1 | Source ↗ |
| Terminal-Bench | harness: Claude Code · variant: high · v2.1 | 64.7 | pass@1 | Source ↗ |
| MMLU-Pro | variant: preview checkpoint; max effort | 87.5 | EM | Source ↗ |
| SimpleQA-Verified | variant: preview checkpoint; max effort | 57.9 | pass@1 | Source ↗ |
| Chinese SimpleQA | variant: preview checkpoint; max effort | 84.4 | pass@1 | Source ↗ |
| GPQA Diamond | variant: preview checkpoint; max effort | 90.1 | pass@1 | Source ↗ |
| Humanity's Last Exam | variant: preview checkpoint; max effort; without tools | 37.7 | pass@1 | Source ↗ |
| LiveCodeBench | variant: preview checkpoint; max effort | 93.5 | pass@1 | Source ↗ |
| Codeforces | variant: preview checkpoint; max effort | 3206 | rating | Source ↗ |
| HMMT | variant: preview checkpoint; max effort · dataset: February 2026 | 95.2 | pass@1 | Source ↗ |
| IMOAnswerBench | variant: preview checkpoint; max effort | 89.8 | pass@1 | Source ↗ |
| MathArena Apex | variant: preview checkpoint; max effort | 38.3 | pass@1 | Source ↗ |
| MathArena Apex Shortlist | variant: preview checkpoint; max effort | 90.2 | pass@1 | Source ↗ |
| MRCR | variant: preview checkpoint; max effort · dataset: 1M context | 83.5 | MMR | Source ↗ |
| CorpusQA | variant: preview checkpoint; max effort · dataset: 1M context | 62 | accuracy | Source ↗ |
| Terminal-Bench | variant: preview checkpoint; max effort · v2.0 | 67.9 | accuracy | Source ↗ |
| SWE-Bench Pro | variant: preview checkpoint; max effort | 55.4 | resolved | Source ↗ |
| SWE-Bench Multilingual | variant: preview checkpoint; max effort | 76.2 | resolved | Source ↗ |
| BrowseComp | variant: preview checkpoint; max effort | 83.4 | pass@1 | Source ↗ |
| Humanity's Last Exam | variant: preview checkpoint; max effort; with tools | 48.2 | pass@1 | Source ↗ |
| MCP Atlas | variant: preview checkpoint; max effort · dataset: public | 73.6 | pass@1 | Source ↗ |
| GDPval-AA | variant: preview checkpoint; max effort | 1554 | Elo | Source ↗ |
| Toolathlon | variant: preview checkpoint; max effort | 51.8 | pass@1 | Source ↗ |
The same benchmark scores very differently across harness / dataset, so the qualifying conditions must be shown together.