DeepSeek V4 Pro vs DeepSeek V4.1 Flash
At a Glance
| Compare |  DeepSeek V4 ProDeepSeek | DeepSeek V4.1 FlashDeepSeek |
|---|---|---|
| Intelligence, Cost, and Efficiency | ||
| IntelligenceHigher is better · MM Intelligence v2.5 | #28 of 4655.3 score · 2/3 sources · provisional · missing ARC-AGI-2 · full-core range 36.9–70.2 | UnrankedNot in the 46-model eligible cohort |
| CostLower is better · Published-token output estimate | #10 of 44$0.054 per LiveBench case | #5 of 44$0.022 per LiveBench case |
| EfficiencyHigher is better · MM Efficiency v1.5 | #12 of 3858.4 score · 2/3 sources · provisional · missing ARC-AGI-2 · full-core range 49.2–65.9 | UnrankedNot in the 38-model eligible cohort |
| Pricing and Limits | ||
| Input priceFrom · USD / 1M tokens | $0.66DeepSeek ↗ · Sep 2, 2026 | $0.15DeepSeek ↗ · Sep 10, 2026 |
| Output priceFrom · USD / 1M tokens | $1.5441Openrouter ↗ · Aug 28, 2026 | $0.60Deepinfra ↗ · Sep 22, 2026 |
| Context windowMaximum documented tokens | 1,049K | 1,049K |
| Model facts checked | Aug 28, 2026View model evidence → | Sep 10, 2026View model evidence → |
Token prices are the lowest available sourced USD rates; input and output may use different providers. Cost ranking estimates output spend on LiveBench, not a full request bill. Ranking methodology →
Available Benchmarks
| Benchmark | DeepSeek-V4-Pro | DeepSeek-V4.1-Flash |
|---|---|---|
| LMArena Agent Arenaagent-2026-09-15-d25aabda0010 · outcome_score · leader | -0.7195% of row best · score · DeepSeek V4 Pro; 95% CI [-1.64679871, 0.21730076]; sessions 35465; observations 1910125; rank 30 | 4.88100% of row best · score · Deepseek V4.1 Flash (Max); 95% CI [3.54852933, 6.21109128]; sessions 20080; observations 2295632; rank 12 |
| LiveBench2026-06-25 · overall · leader | 76.7992% of row best · percent · deepseek-v4-pro · 35,014 output tokens / case | 83.20100% of row best · percent · deepseek-v4.1-flash-max · 36,355 output tokens / case |
| ToneBench2026-09-11-10-task-4ef099199c9c · overall_score · unverified | 79.2590% of row best · points · DeepSeek V4 Pro (xhigh) · 9,652 output tokens / case | 88.03100% of row best · points · DeepSeek V4.1 Flash (max) · 18,199 output tokens / case |
| Overall ResultCounted from the protocol-matched rows above | 0 benchmark wins | 2 benchmark winsOverall lead |
Third-party benchmark Only like-for-like primary-publisher results are shown; raw scores, relative scores, configuration, and token spend remain visible.
Quality Versus Estimated Output Cost
Side-by-Side Facts
| Field | DeepSeek-V4-Pro | DeepSeek-V4.1-Flash |
|---|---|---|
| Developer | DeepSeek | DeepSeek |
| Family | Deepseek V4 Pro | Deepseek V4 1 |
| Model | DeepSeek-V4-Pro | DeepSeek-V4.1-Flash |
| Version | DeepSeek-V4-Pro | DeepSeek-V4.1-Flash |
| Lifecycle | active | active |
| Released | 2026-04-24 | 2026-09-10 |
| Knowledge cutoff | Unknown | Unknown |
| Input modalities | Text | Text, Image |
| Output modalities | Text | Text |
| Context window | 1,049K | 1,049K |
| Total parameters | 1.6T | 763.2B |
| Active parameters | 49B | Unknown |
| License | mit | mit |
| Open weights | Yes | Yes |
| API available | Yes | Yes |
| Self-hostable | Yes | Yes |
| Provider access | DeepSeek (Standard), Deepinfra (Standard), Fireworks Ai (Standard), Hugging Face (Standard), Openrouter (Standard), Together Ai (Standard) | DeepSeek (Standard), Deepinfra (Standard), Together Ai (Standard) |
| Capabilities | chat, generation, reasoning | agents, chat, fim, generation, reasoning, responses, structured_outputs, tools, vision |
| Architecture design | Unknown | Causal Encoder-Decoder (20 encoder + 20 decoder layers) |
| Backbone parameters | Unknown | 552000000000 parameters |
| Active parameters during decode | Unknown | 16000000000 parameters |
| Active parameters during prefill | Unknown | 8000000000 parameters |
| Pre-training corpus | Unknown | 45000000000000 tokens |
| Reasoning effort range | Unknown | 1–100 |
| Routed experts per MoE layer | Unknown | 384 experts |
| Routed experts per token | Unknown | 6 experts |
| Transformer layers | Unknown | 40 layers |
DeepSeek V4 Pro Capabilities
DeepSeek V4.1 Flash Capabilities
Related Comparisons
All text comparisons →| A | Pair | B | Context |
|---|---|---|---|
DeepSeek-V3.2DeepSeek | vs | DeepSeek-V4-ProDeepSeek | family variantstext |
DeepSeek-V4-FlashDeepSeek | vs | DeepSeek-V4-ProDeepSeek | family variantstext |
DeepSeek-V4-ProDeepSeek | vs |  GPT-6 AstraOpenAI | cross-developer peerstext |
Claude Fable 5.1Anthropic | vs | DeepSeek-V4-ProDeepSeek | cross-developer peerstext |
DeepSeek-V4-ProDeepSeek | vs |  Gemini 3.1 ProGoogle DeepMind | cross-developer peerstext |
DeepSeek-V4-ProDeepSeek | vs | Grok 4.6xAI | cross-developer peerstext |
DeepSeek-V4-ProDeepSeek | vs |  Qwen3.8-MaxQwen | cross-developer peerstext |
DeepSeek-V4-ProDeepSeek | vs |  Kimi-K3Moonshot AI | cross-developer peerstext |
 MiniMax-M3MiniMax | vs | DeepSeek-V4-ProDeepSeek | cross-developer peerstext |
DeepSeek-V4-ProDeepSeek | vs |  GLM-5.3Z.ai | cross-developer peerstext |
DeepSeek-V4-ProDeepSeek | vs |  Hy4 previewTencent | cross-developer peerstext |
Seed 2.1 ProByteDance Seed | vs | DeepSeek-V4-ProDeepSeek | cross-developer peerstext |
Primary Evidence
Sources and Freshness
Questions
DeepSeek V4 Pro vs DeepSeek V4.1 Flash FAQs
Is DeepSeek V4 Pro or DeepSeek V4.1 Flash better for coding?+
This comparison does not currently contain a protocol-matched coding benchmark for both DeepSeek V4 Pro and DeepSeek V4.1 Flash, so Model Markets cannot name a coding leader from pricing, context size, or capability labels alone.
Which is cheaper, DeepSeek V4 Pro or DeepSeek V4.1 Flash?+
DeepSeek V4 Pro is $0.66 and DeepSeek V4.1 Flash is $0.15 per million tokens, so DeepSeek V4.1 Flash is cheaper on this metric. DeepSeek V4 Pro is $1.5441 and DeepSeek V4.1 Flash is $0.60 per million tokens, so DeepSeek V4.1 Flash is cheaper on this metric.
Which has a larger context window, DeepSeek V4 Pro or DeepSeek V4.1 Flash?+
Neither model has a larger sourced context window in this comparison. DeepSeek V4 Pro is 1,049K and DeepSeek V4.1 Flash is 1,049K.
Which performs better in benchmarks, DeepSeek V4 Pro or DeepSeek V4.1 Flash?+
DeepSeek V4.1 Flash leads the current overall benchmark count. The result uses 2 protocol-matched benchmarks from 2 publishers; it is not a universal quality score.
Can DeepSeek V4 Pro or DeepSeek V4.1 Flash be self-hosted?+
Both models have the same recorded self-hosting status: supported. DeepSeek V4 Pro is open weight; DeepSeek V4.1 Flash is open weight.
Can DeepSeek V4 Pro and DeepSeek V4.1 Flash understand images?+
DeepSeek V4 Pro is not documented with image input; DeepSeek V4.1 Flash is documented with image input. This reflects supported input modalities, not vision quality.
Which can generate longer answers, DeepSeek V4 Pro or DeepSeek V4.1 Flash?+
Neither has a larger sourced maximum output. DeepSeek V4 Pro is — and DeepSeek V4.1 Flash is 393K.
Do DeepSeek V4 Pro and DeepSeek V4.1 Flash support reasoning and tool use?+
DeepSeek V4 Pro: reasoning. DeepSeek V4.1 Flash: reasoning, tool calling, and image input. Feature support does not establish relative quality.
Which is available from more inference providers, DeepSeek V4 Pro or DeepSeek V4.1 Flash?+
DeepSeek V4 Pro has 6 sourced provider routes; DeepSeek V4.1 Flash has 3, so DeepSeek V4 Pro has broader tracked availability.
Which offers better value, DeepSeek V4 Pro or DeepSeek V4.1 Flash?+
There is no universal value winner. Compare the input and output prices above with the matched benchmark result for your workload: cheaper tokens can be offset by different quality, token usage, latency, or provider availability.