Text and chat · DeepSeek
DeepSeek
Low-cost open models with 1M context and good reasoning; V4.1 Flash adds image reading. Pick it to do a lot for very little.
Private Low costWhat sets it apart
DeepSeek competes on price and speed: its Flash tier is a 552B mixture of experts with only 16B active parameters, a 1M context window and MIT weights, at $0.30 per million input tokens. The Pro tier goes to 1.6T parameters for harder problems.
- Flash tier at $0.30 per million input
- 1M context, MIT weights
- Pro tier: 1.6T MoE, 49B active
Recommended for: work over a very long document or many notes at once.
Strengths and limits
Where it shines
- Very low price per token, private mode
- 1M token context on every V4 version
- Tool calling and adjustable reasoning effort
Think twice when
- Only V4.1 Flash reads images, and its maker says it lags on complex images
- V4.1 Flash is the newest version: prefer it for new work
- 1M token context (V3.2: 160K)
- Vision on V4.1 Flash only
- Open weights
Use cases
High-volume summaries on a budget
Summarize each of these 40 customer emails in two sentences and tag it as billing, bug or feature request.
V4.1 Flash costs $0.30 per million input tokens and answers fast.
Working through a hard problem
Here are the constraints for our delivery planning. Find a schedule that satisfies all of them and explain why it works.
The Pro tier is a reasoning model with 1.6T parameters.
Reading a very long file
Read this 400-page report and list the five decisions it asks the board to make.
Every V4 model takes a 1M token context.
Against its rivals
- GLMGLM 5.3 scores higher on the index (45 vs 39) but costs more per token.#13 of 90 on Artificial Analysis Intelligence Index (45 points)
- Qwen open modelsQwen’s open line spans many sizes down to phones; DeepSeek stays at large sizes.#23 of 90 on Artificial Analysis Intelligence Index (40 points)
- KimiKimi K3 reads images and ranks near GLM, but at several times DeepSeek’s price.#15 of 90 on Artificial Analysis Intelligence Index (44 points)
Version history
Each version, dated, with what it changed over the one before.
Hollow points are earlier versions no longer offered in Sub Rosa.
See the numbers
| Version | Released | points |
|---|---|---|
| DeepSeek V3.2 | 2025-12-01 | 16 |
| DeepSeek V4 Flash 0731 | 2026-07-31 | 34 |
| DeepSeek V4 Pro 0813 | 2026-08-13 | 36 |
| DeepSeek V4.1 Flash | 2026-09-10 | 39 |
- DeepSeek V3.2
- 685B mixture of experts (37B active) with a 128K context, MIT license.
- DeepSeek V4 (April 2026)
- V4 generation: the Pro model has 1.6T parameters (49B active) and a 1M token context, up from 128K.
- DeepSeek V4 Flash 0731
- Flash tier of 284B parameters with 13B active, 1M context, text only, scoring 34 on the Artificial Analysis index.
- DeepSeek V4 Pro 0813
- Refreshed 1.6T Pro model, scoring 36 on the index against 30 for the April Pro.
- DeepSeek V4.1 FlashLatest
- Larger Flash: 552B parameters (16B active), now accepts images, and cuts cache memory to about a quarter via a new attention design.
Benchmarks
How to read these scores: what a benchmark tells you.
Artificial Analysis Intelligence Index: #24 of 90 models measured in the catalog. One score averaging about ten independent tests of reasoning, knowledge, math, coding and agent work, all run the same way for every model. Higher is better. Each model is listed at its highest reasoning effort. Two or three points apart is a close call; ten points is a different class of model.
See the numbers
| Model | points |
|---|---|
| 1. Claude Opus 5.5 | 58 |
| 2. Claude Sonnet 5.5 | 56 |
| 3. Claude Fable 5.1 | 53 |
| 4. GPT-6 Astra | 53 |
| 5. Gemini 4 Argon | 53 |
| 6. GPT-6.1 Sol | 52 |
| 7. Muse Wonder | 52 |
| 8. Claude Opus 5 | 51 |
| 9. Claude Fable 5 | 50 |
| 10. GPT-6 Sol | 48 |
| 24. DeepSeek V4.1 Flash | 39 |
| 34. DeepSeek V4 Pro 0813 | 36 |
| 36. DeepSeek V4 Flash 0731 | 34 |
Announced by the maker
Measured by the maker with its own settings: useful, but not directly comparable with the independent scores above.
| Benchmark | Model | Score | Source |
|---|---|---|---|
| gpqa-diamond | DeepSeek V4.1 Flash | 90.9 | huggingface.co |
All 8 models in this family
| Model | Use | Privacy | Price |
|---|---|---|---|
DeepSeek V4.1 Flashdeepseek-v4-1-flash | Text | Private | $0.54 per million tokens written$0.000378 a page written |
DeepSeek V4 Flash 0423deepseek-v4-flash | Text | Private | $0.124 per million tokens written$0.0000865 a page written |
DeepSeek V4 Flash 0731deepseek-v4-flash-0731 | Text | Private | $0.117 per million tokens written$0.0000818 a page written |
DeepSeek V4 Flash 0731 Fastdeepseek-v4-flash-0731-fast | Text | Private | $0.315 per million tokens written$0.00022 a page written |
DeepSeek V4 Prodeepseek-v4-pro | Text | Private | $1.484 per million tokens written$0.00104 a page written |
DeepSeek V4 Pro 0813deepseek-v4-pro-0813 | Text | Private | $2.226 per million tokens written$0.00156 a page written |
DeepSeek V3.2deepseek-v3.2 | Text | Private | $0.216 per million tokens written$0.000151 a page written |
DeepSeek V4 Flash E2EEe2ee-deepseek-v4-flash | Text | Private | $0.168 per million tokens written$0.000117 a page written |
Specifications
| Specification | DeepSeek V4.1 Flash | DeepSeek V4 Pro 0813 | DeepSeek V4 Flash 0731 | DeepSeek V3.2 |
|---|---|---|---|---|
| Released | September 10, 2026 | August 13, 2026 | July 31, 2026 | December 1, 2025 |
| Parameters | 552B total, 16B active | 1.6T total, 49B active | Not disclosed | 685B total, 37B active |
| Weights | Open, MIT | Open, MIT | Open, MIT | Open, MIT |
| Accepts | text, images | text | text | text |
| Produces | text | text | text | text |
| Context | 1,000K tokens | 1,000K tokens | 1,000K tokens | 160K tokens |
| Abilities | Reads images, Uses Sub Rosa’s tools, Thinks before answering, Tuned for code | Uses Sub Rosa’s tools, Thinks before answering, Tuned for code | Uses Sub Rosa’s tools, Thinks before answering, Tuned for code | Uses Sub Rosa’s tools, Thinks before answering |
| Price | $0.54 per million tokens written | $2.226 per million tokens written | $0.117 per million tokens written | $0.216 per million tokens written |
| Privacy | Private | Private | Private | Private |
Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.
Sources
Everything on this page comes from these pages, read on October 16, 2026 or as dated on each score.
Show the 9 sources
- artificialanalysis.ai/models/deepseek-v4-1-flash
- artificialanalysis.ai/models/deepseek-v4-pro
- artificialanalysis.ai/models/deepseek-v4-pro-0424
- artificialanalysis.ai/models/deepseek-v4-flash-0731
- artificialanalysis.ai/models/deepseek-v3-2
- huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash
- artificialanalysis.ai/leaderboards/models
- thenextweb.com/news/deepseek-v4-1-flash-launch-v4-pro-retired-price-cut
- propakistani.pk/2026/09/11/deepseek-v4-1-flash-launches-with-lower-prices-and-native-vision