DeepSeek · language model
DeepSeek V3.2 Exp: pricing, context and benchmarks
DeepSeek V3.2 Exp is DeepSeek's model released on September 29, 2025. It costs $0.27 per million input tokens and $0.41 per million output tokens, with a 163.8K-token context window. Epoch AI ranks it #85 of 125 rated models.
What you need to know about DeepSeek V3.2 Exp
DeepSeek, based in China, put DeepSeek V3.2 Exp online on September 29, 2025. It accepts text as input and produces text. Its context window reaches 163,840 tokens, larger than 32 % of the models in the registry, and it can write up to 147,456 tokens in one go.
It has been available for about 13 months, long enough for its prices and scores to have settled.
DeepSeek V3.2 Exp pricing
On OpenRouter, DeepSeek V3.2 Exp costs $0.27 per million input tokens and $0.41 per million output tokens. Counting three tokens read for every token written, its average price is $0.305 per million, more expensive than 29 % of paid models (2.5 times cheaper than the median of $0.76).
A concrete example: a task that sends 2,000 tokens (about 1,500 words) and receives a 500-token answer costs $0.0007. At 1,000 tasks a day, budget $22 a month. The price calculator redoes this with your own volumes.
Performance and benchmarks
The Epoch Capabilities Index (ECI), which aggregates published results across many benchmarks, gives DeepSeek V3.2 Exp a score of 144.9 (range 142.5 to 146.8). That places it #85 of 125 rated models.
Scores published by Epoch AI on the benchmarks the site tracks:
| Benchmark | What it measures | Score | Date |
|---|---|---|---|
| Aider Polyglot | coding exercises in several languages | 74.2 % | Sep 29, 2025 |
For at least equal performance, Claude Haiku 5.5, GPT-6 Luna, and DeepSeek V4 Flash 0731 cost at least 30% less.
Features, privacy and local use
- It supports tool calling, so it can drive an AI agent, call APIs or MCP servers.
- It can answer in an enforced JSON format (structured outputs).
- It is a reasoning model: it "thinks" before answering, which helps on hard tasks but uses more output tokens.
- Its knowledge stops at July 2025.
Its weights are open and downloadable from Hugging Face (mit licence): you can run it on your own machines without sending your data to a third party. With 685,396,921,376 billion parameters, it needs about 382,995,375,270 GB of memory at 4-bit quantisation. Check with the VRAM calculator. It was downloaded 40,373 times on Hugging Face last month.
Other DeepSeek models
| Model | Released | Avg. price | ECI |
|---|---|---|---|
| DeepSeek V4.1 Flash 1017 | Oct 17, 2026 | $0.26 | 155.9 |
| DeepSeek V4.1 Flash | Sep 10, 2026 | $0.27 | 155.9 |
| DeepSeek V4 Flash Vision Exp | Aug 21, 2026 | $0.32 | — |
| DeepSeek V4 Pro 0813 | Aug 12, 2026 | $0.99 | 155.6 |
| DeepSeek V4 Flash 0731 | Jul 31, 2026 | $0.07 | 154.8 |
| DeepSeek V4 Pro 0423 | Apr 24, 2026 | $0.26 | 155.6 |
| DeepSeek V4 Flash 0423 | Apr 24, 2026 | $0.11 | 154.8 |
| DeepSeek V3.2 | Dec 1, 2025 | $0.32 | 146.1 |
Verdict
Computed score: 55/100, from public data (performance, price, context, features).
Strengths
- Low price: cheaper than 71 % of paid models.
- Open weights: self-hostable, your data stays with you.
- Agent-ready: tool calling and structured JSON output.
Limitations
- Text only: reads neither images nor files.
Avoid if…
- you want to host it on a single consumer computer (≈ 382,995,375,270 GB needed at 4-bit).
Alternatives
- MiniMax M2.7$0.53
- Qwen3.6 35B A3B$0.31
- gpt-oss-120b$0.26
- Qwen3.5-35B-A3B$0.36
- Gemma 4 31B$0.25
- MiniMax M2.5$0.53
Comparisons and tools
Learn more
Frequently asked questions
How much does DeepSeek V3.2 Exp cost?
$0.27 per million input tokens and $0.41 per million output tokens on OpenRouter (checked October 25, 2026). A task with 2,000 tokens read and 500 written costs $0.0007.
How big is DeepSeek V3.2 Exp's context window?
163,840 tokens, roughly 123,000 words, with a maximum of 147,456 output tokens.
Can DeepSeek V3.2 Exp run locally?
Yes, its weights are published under the mit licence. Expect about 382,995,375,270 GB of memory at 4-bit.
Is DeepSeek V3.2 Exp suitable for an AI agent?
Technically yes: it supports tool calling and structured outputs. Then check its cost for your volume.
How good is DeepSeek V3.2 Exp?
Epoch AI ranks it #85 of 125 rated models (ECI 144.9). On Aider Polyglot, it scores 74.2%.
What are the alternatives to DeepSeek V3.2 Exp?
The closest models on price and performance are MiniMax M2.7, Qwen3.6 35B A3B, and gpt-oss-120b.
Related articles
- Mistral AI: models, pricing and agentsOctober 25, 2026
- LLM benchmarks: how to read model rankingsOctober 24, 2026
- Running a local LLM: hardware, software, and limitationsOctober 24, 2026
- Open-source or closed models for your AI agents?October 5, 2026
Neutrality. No affiliate links. No vendor pays to be reviewed, rated or ranked. Reviews are directed by François Chollet, neutral and factual, and every data point is dated and sourced. How we score