DeepSeek V3.2 vs DeepSeek V4 Pro
At a Glance
| Compare |  DeepSeek V3.2DeepSeek | DeepSeek V4 ProDeepSeek |
|---|---|---|
| Intelligence, Cost, and Efficiency | ||
| IntelligenceHigher is better · MM Intelligence v4.4 | UnrankedNot in the 57-model eligible cohort | #19 of 5757.9 score · 2/3 sources · provisional · missing ARC-AGI-2 · full-core range 38.6–72.0 |
| CostLower is better · Published-token output estimate | UnrankedNot in the 27-model eligible cohort | #10 of 27$0.054 per LiveBench case |
| EfficiencyHigher is better · MM Efficiency v1.9 | UnrankedNot in the 27-model eligible cohort | #11 of 2751.5 score · 2/3 sources · provisional · missing ARC-AGI-2 · full-core range 41.9–58.5 |
| Pricing and Limits | ||
| Input priceFrom · USD / 1M tokens | $0.26Deepinfra ↗ · Oct 19, 2026 | $0.66DeepSeek ↗ · Sep 2, 2026 |
| Output priceFrom · USD / 1M tokens | $0.38Deepinfra ↗ · Oct 19, 2026 | $1.5441Openrouter ↗ · Aug 28, 2026 |
| Context windowMaximum documented tokens | 164K | 1,049K |
| Model facts checked | Aug 28, 2026View model evidence → | Aug 28, 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-V3.2 | DeepSeek-V4-Pro |
|---|---|---|
| LMArena Text Arenatext-2026-10-15-a64ca3b12b6e · arena_rating · leader | 1,424.4698% of row best · rating · deepseek-v3.2; 95% CI [1420.99052972, 1427.92299771]; votes 48112; rank 105 | 1,451.16100% of row best · rating · deepseek-v4-pro; 95% CI [1447.27741116, 1455.03943137]; votes 57594; rank 49 |
| Overall ResultCounted from the protocol-matched rows above | 0 benchmark winsNo overall winner | 1 benchmark winNo overall winner |
Third-party benchmark Only like-for-like primary-publisher results are shown; raw scores, relative scores, configuration, and token spend remain visible.
Side-by-Side Facts
| Field | DeepSeek-V3.2 | DeepSeek-V4-Pro |
|---|---|---|
| Developer | DeepSeek | DeepSeek |
| Family | Deepseek V3 2 | Deepseek V4 Pro |
| Model | DeepSeek-V3.2 | DeepSeek-V4-Pro |
| Version | DeepSeek-V3.2 | DeepSeek-V4-Pro |
| Lifecycle | active | active |
| Released | 2025-12-01 | 2026-04-24 |
| Knowledge cutoff | Unknown | Unknown |
| Input modalities | Text | Text |
| Output modalities | Text | Text |
| Context window | 164K | 1,049K |
| Total parameters | 685.4B | 1.6T |
| Active parameters | Unknown | 49B |
| License | mit | mit |
| Open weights | Yes | Yes |
| API available | Yes | Yes |
| Self-hostable | Yes | Yes |
| Provider access | Deepinfra (Standard), Hugging Face (Standard), Openrouter (Standard) | DeepSeek (Standard), Deepinfra (Standard), Fireworks Ai (Standard), Hugging Face (Standard), Openrouter (Standard), Together Ai (Standard) |
| Capabilities | chat, generation, reasoning | chat, generation, reasoning |
DeepSeek V3.2 Capabilities
DeepSeek V4 Pro Capabilities
Related Comparisons
All text comparisons →| A | Pair | B | Context |
|---|---|---|---|
DeepSeek-V3.2DeepSeek | vs |  Phi-4-reasoningMicrosoft | cross-developer peerstext |
DeepSeek-V3.2DeepSeek | vs | DeepSeek-V4-Pro-BaseDeepSeek | family variantstext |
DeepSeek-R1DeepSeek | vs | DeepSeek-V3.2DeepSeek | family variantstext |
DeepSeek-V3DeepSeek | vs | DeepSeek-V3.2DeepSeek | family variantstext |
DeepSeek-V3.2DeepSeek | vs |  Llama-3.1-405BMeta | cross-developer peerstext |
DeepSeek-V3.2DeepSeek | vs | cross-developer peerstext | |
DeepSeek-V3.2DeepSeek | vs | Llama-3.1-70BMeta | cross-developer peerstext |
DeepSeek-V3.2DeepSeek | vs |  Kimi-K2-ThinkingMoonshot AI | cross-developer peerstext |
DeepSeek-V3.2DeepSeek | vs |  GLM-5Z.ai | cross-developer peerstext |
DeepSeek-V3.2DeepSeek | vs | GLM-5.1Z.ai | cross-developer peerstext |
DeepSeek-V3.2DeepSeek | vs | GLM-5.2Z.ai | cross-developer peerstext |
Primary Evidence
Sources and Freshness
Questions
DeepSeek V3.2 vs DeepSeek V4 Pro FAQs
Is DeepSeek V3.2 or DeepSeek V4 Pro better for coding?+
This comparison does not currently contain a protocol-matched coding benchmark for both DeepSeek V3.2 and DeepSeek V4 Pro, so Model Markets cannot name a coding leader from pricing, context size, or capability labels alone.
Which is cheaper, DeepSeek V3.2 or DeepSeek V4 Pro?+
DeepSeek V3.2 is $0.26 and DeepSeek V4 Pro is $0.66 per million tokens, so DeepSeek V3.2 is cheaper on this metric. DeepSeek V3.2 is $0.38 and DeepSeek V4 Pro is $1.5441 per million tokens, so DeepSeek V3.2 is cheaper on this metric.
Which has a larger context window, DeepSeek V3.2 or DeepSeek V4 Pro?+
DeepSeek V4 Pro has the larger sourced context window. DeepSeek V3.2 supports 163,840 tokens and DeepSeek V4 Pro supports 1,048,576 tokens.
Which performs better in benchmarks, DeepSeek V3.2 or DeepSeek V4 Pro?+
There is no overall benchmark winner: At least two independently verified, protocol-matched benchmarks are required for an overall winner.
Can DeepSeek V3.2 or DeepSeek V4 Pro be self-hosted?+
Both models have the same sourced self-hosting status: supported. DeepSeek V3.2 is open weight; DeepSeek V4 Pro is open weight.
Can DeepSeek V3.2 and DeepSeek V4 Pro understand images?+
DeepSeek V3.2 is not documented with image input; DeepSeek V4 Pro is not documented with image input. This reflects supported input modalities, not vision quality.
Which can generate longer answers, DeepSeek V3.2 or DeepSeek V4 Pro?+
Neither model has a sourced maximum output in this comparison, so a larger limit cannot be determined.
Do DeepSeek V3.2 and DeepSeek V4 Pro support reasoning and tool use?+
DeepSeek V3.2: reasoning. DeepSeek V4 Pro: reasoning. Feature support does not establish relative quality.
Which is available from more inference providers, DeepSeek V3.2 or DeepSeek V4 Pro?+
DeepSeek V3.2 has 3 sourced provider routes; DeepSeek V4 Pro has 6, so DeepSeek V4 Pro has broader tracked availability.
Which offers better value, DeepSeek V3.2 or DeepSeek V4 Pro?+
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.