How much does DeepSeek-V3.2 cost?

DeepSeek-V3.2 costs $0.28 per million input tokens and $0.40 per million output tokens, using the list price from DeepSeek as of 26 Oct 2026. That is about $0.05 for 100 chats of about 1,000 tokens in and 500 tokens back. Confirm on the maker's site before you buy, because prices change.

What is the context window of DeepSeek-V3.2?

DeepSeek-V3.2 has a context window of 164K tokens, roughly 123,000 words of text read at once. A single answer can be up to 164K tokens long.

Is DeepSeek-V3.2 open source?

DeepSeek-V3.2 has open weights: its trained parameters can be downloaded. That is not always the same as open source, because licence terms differ, so read the licence before you use it commercially.

How good is DeepSeek-V3.2 compared with other AI models?

On the Epoch Capabilities Index, DeepSeek-V3.2 scores 146.1 (likely range 144.3 to 147.4), which ranks 82 of 270 models we track. On Arena's human-vote text board it is ranked 86 of 166. Scores close together are a tie in practice.

How does DeepSeek-V3.2 compare with GPT-5.4 Nano?

DeepSeek-V3.2 scores higher on 3 of 6 shared quality measures, GPT-5.4 Nano on 0. 3 measures are too close to call. DeepSeek-V3.2 costs about 33% less than GPT-5.4 Nano on a blended price of $0.31 against $0.46 per million tokens. See the full comparison for every measure.

How current are these numbers for DeepSeek-V3.2?

The newest data on this page is dated 26 Oct 2026. Scores come from Epoch AI and Arena, prices from the open LiteLLM price list.

AI Rankings

DeepSeek-V3.2: Score, Price and Context Window

Where DeepSeek-V3.2 from DeepSeek ranks, what it costs and how much text it can read, in plain numbers with the sources.

Data updated 26 Oct 2026

DeepSeek
Open weights
Released Dec 2025

Its score range overlaps with DeepSeek-V4-Flash, the model just above it, so a small gap in rank is not a meaningful difference.

Best for

Keeping costs low: it is among the cheapest quarter of models in the top half by score.
Running it yourself: its weights can be downloaded (read the licence terms first).

These lines are worked out from the numbers on this page, not from hands-on testing.

Key facts

Overall score: 146.1; Rank 82 of 270. Likely range 144.3 to 147.4.
Human preference: chat: 1425; Rank 86 of 166. 48,090 votes.
Human preference: coding: 1476; Rank 83 of 166. 8,880 votes, at the "Thinking" setting.
Human preference: web pages: 1363; Rank 76 of 85. 6,925 votes, at the "Thinking" setting.
Price per million tokens: $0.28 in / $0.40 out; Blended $0.31. Cached input $0.028. List price from DeepSeek, checked 26 Oct 2026.
Context window: 164K tokens; About 123,000 words at once. One answer up to 164K tokens.
Weights: Open; The weights can be downloaded. Check the licence before commercial use.

Features listed by the price list: Tool use, Reasoning, Prompt caching.

Try itDeepSeek in our AI tools directoryAI assistant and model platform for reasoning, coding, writing, and conversational work, with separately billed APIs.

How DeepSeek-V3.2 stacks up

Each strip shows every model we track as faint bars. The dot is this model. Better is always to the right, even for price.

DeepSeek-V3.2 compared with all models we trackOverall capability: 146.1. Better than 69% of the other 269 models. Human preference: chat: 1425. Better than 48% of the other 165 models. Human preference: coding: 1476. Better than 50% of the other 165 models. Human preference: web pages: 1363. Better than 10% of the other 84 models. Price: $0.31 per 1M. Cheaper than 75% of the other 200 priced models. Context window: 164K tokens. Larger context window than 36% of the other 198 models.

Overall capability146.1
Better than 69% of the other 269 modelsUpper half
Epoch Capabilities Index: one score that combines many tests.
Human preference: chat1425
Better than 48% of the other 165 models
How often people pick its answers over another model's on Arena.
Human preference: coding1476
Better than 50% of the other 165 modelsUpper half
How often people pick its answers to coding questions on Arena.
Human preference: web pages1363
Better than 10% of the other 84 models
How often people pick the web pages and small apps it builds on Arena.
Price$0.31 per 1M
Cheaper than 75% of the other 200 priced modelsTop 25%
Blended price per million tokens. Lower is better, so the cheapest models sit on the right.
Context window164K tokens
Larger context window than 36% of the other 198 models
How much text it can read at once. Bigger suits long documents.

Compared with the other published models that have each measure. Data updated 26 Oct 2026.

Where DeepSeek-V3.2 ranks

Leaderboards it appears on, best placing first.

See all leaderboards

Test results

Each test checks one kind of skill. Place shows where it stands among the models that took that test.

Aider Polyglot
225 coding exercises in six programming languages. The AI edits the code files and the exercise's tests decide pass or fail.
74.2%
Place 7 of 46
tied with 1 other
OTIS Mock AIME
Maths competition problems in the style of the US AIME exam, from the 2024 and 2025 sets. Every answer is a whole number.
87.8%
Place 65 of 174
GPQA Diamond
Very hard multiple-choice science questions in biology, physics and chemistry, written by experts and designed so a quick web search does not give the answer.
83.4%
Place 75 of 183
Terminal-Bench
Tasks done in a computer terminal, such as setting up software or fixing a broken build, checked by automatic tests. Scores here use the benchmark's standard reference agent so models are compared on the same footing.
39.5%
Place 12 of 25
ARC-AGI-2
Visual puzzles where the AI must work out a new rule from a few examples and apply it. They test picking up a new idea quickly, not recalling facts.
4.0%
Place 64 of 86
tied with 1 other

What it costs in everyday terms

List prices turned into amounts you can picture. Real bills depend on how you use it.

100 chats: $0.05; Cost of 100 chats assumes each chat sends 1,000 tokens (about 750 words) and gets 500 tokens (about 375 words) back.
A 300-page book: $0.03; Cost to read a 300-page book assumes about 100,000 input tokens and no reply. A real book varies with page size and language.
Blended price: $0.31 per 1M; Three parts input to one part output. Our own calculation for comparing models.

Similar models

Models from other makers with a similar overall score.

Learn more

Questions about DeepSeek-V3.2

How much does DeepSeek-V3.2 cost?: DeepSeek-V3.2 costs $0.28 per million input tokens and $0.40 per million output tokens, using the list price from DeepSeek as of 26 Oct 2026. That is about $0.05 for 100 chats of about 1,000 tokens in and 500 tokens back. Confirm on the maker's site before you buy, because prices change.
What is the context window of DeepSeek-V3.2?: DeepSeek-V3.2 has a context window of 164K tokens, roughly 123,000 words of text read at once. A single answer can be up to 164K tokens long.
Is DeepSeek-V3.2 open source?: DeepSeek-V3.2 has open weights: its trained parameters can be downloaded. That is not always the same as open source, because licence terms differ, so read the licence before you use it commercially.
How good is DeepSeek-V3.2 compared with other AI models?: On the Epoch Capabilities Index, DeepSeek-V3.2 scores 146.1 (likely range 144.3 to 147.4), which ranks 82 of 270 models we track. On Arena's human-vote text board it is ranked 86 of 166. Scores close together are a tie in practice.
How does DeepSeek-V3.2 compare with GPT-5.4 Nano?: DeepSeek-V3.2 scores higher on 3 of 6 shared quality measures, GPT-5.4 Nano on 0. 3 measures are too close to call. DeepSeek-V3.2 costs about 33% less than GPT-5.4 Nano on a blended price of $0.31 against $0.46 per million tokens. See the full comparison for every measure.
How current are these numbers for DeepSeek-V3.2?: The newest data on this page is dated 26 Oct 2026. Scores come from Epoch AI and Arena, prices from the open LiteLLM price list.

Where this data comes from

Epoch AI: Benchmark data: Epoch AI, 'Capabilities & benchmarking', epoch.ai/benchmarks (CC BY 4.0). We match models across sources, rank them and show them in charts. Scores are not changed. Epoch AI does not endorse this site.
Arena (formerly LMArena): Human-preference ratings: Arena (arena.ai) leaderboard dataset, CC BY 4.0. We keep the latest published ratings, match them to models from other sources and rank them. Ratings are not changed. Arena does not endorse this site.
LiteLLM model price list: Prices and context windows: LiteLLM model price list (github.com/BerriAI/litellm, MIT licence). We use the model maker's own price entry when there is one. Prices change often, so always confirm on the maker's site before you buy.

Newest data on this page: 26 Oct 2026. Prices are list prices and change often, so confirm on the maker's site before you buy. Sponsors never change a score or an order. How we calculate this · Report a mistake