# DeepSeek V4.1 Flash: 1M context, MIT weights, $0.15 per 1M input

> DeepSeek's current Flash model reads images and text, handles 1M tokens and is MIT-licensed. The latest update, Flash-1017, shipped on October 17, 2026.

URL: https://claudegpt.si/news/deepseek-v4-1-flash-explained  
Published: 2026-10-29T18:51:27.695Z  
Author: ClaudeGPT Si Newsroom  
Section: AI Models  
Topics: DeepSeek, Open-weight models, API pricing, Hugging Face  
Publisher: ClaudeGPT Si

## Key takeaways

- DeepSeek released the DeepSeek-V4.1-Flash-1017 update on October 17, 2026, and the API name deepseek-flash now points to it.
- It has a 1M-token context window, takes images and text, and its weights are on Hugging Face under the MIT license.
- As of October 29, 2026, API input costs $0.15 per 1M tokens on a cache miss off-peak, and output costs $0.60.

DeepSeek V4.1 Flash is DeepSeek's current Flash model: a multimodal, open-weight mixture-of-experts model that reads images and text, handles 1 million tokens of context and is released under the MIT license. The version most people mean is [DeepSeek-V4.1-Flash-1017](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017), a refresh DeepSeek shipped on October 17, 2026. Its API name is `deepseek-flash`, and it costs $0.15 per million input tokens off-peak, as of October 29, 2026.

A mixture-of-experts model runs only part of its network for each token, which is how a model this large stays cheap to serve. Most of the searching right now is about which model is which, because DeepSeek changed the names. This is what we could confirm from DeepSeek's own pages.

## What DeepSeek V4.1 Flash is

There are two Hugging Face repositories with nearly the same name. `DeepSeek-V4.1-Flash` is the base release of the model. `DeepSeek-V4.1-Flash-1017` is a minor update to it, which the model card says aims at better reasoning, coding, agent and visual performance, with faster task completion and shorter outputs. DeepSeek's [release note](https://api-docs.deepseek.com/news/news261017) dated October 17, 2026 describes the update as a Pareto improvement in performance and cost, meaning better results without a higher price.

Details from the model cards:

- It takes images and text as input and writes text.
- The base card lists 552B backbone parameters, with a total of 763B in the header. It says about 8B parameters are active per token during prefill and 16B during decode.
- The context window is 1M tokens. The card recommends allowing at least 256K tokens of output.
- The base card says its global KV cache, the memory a model keeps for the text it has read, is about a quarter of V4-Flash's per token. It also adds adjustable reasoning effort from 1 to 100.

The older `DeepSeek-V4-Flash` is a smaller model, with a 284B backbone according to the V4.1 card. The V4.1 generation is a different, larger design, not a point release of it.

## Price, context and API name

DeepSeek's [pricing page](https://api-docs.deepseek.com/quick_start/pricing) lists these prices per 1M tokens in US dollars, as of October 29, 2026:

| Per 1M tokens | Off-peak | Peak |
| --- | --- | --- |
| Input, cache hit | $0.003 | $0.006 |
| Input, cache miss | $0.15 | $0.30 |
| Output | $0.60 | $1.20 |

Context length is 1M tokens and the maximum output is 384K. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, except Chinese public holidays. DeepSeek says it may change prices. Our [DeepSeek pricing article](/news/is-deepseek-free) covers the full price table, the V4 Pro model and the free chat app.

To call it, use `deepseek-flash` as the model name. The docs say the calling method is unchanged. The older names `deepseek-v4-flash` and `deepseek-v4-flash-vision-exp` still work, but those models are retired: requests to them are served by DeepSeek-V4.1-Flash-1017 and billed at the Flash price.

## Is DeepSeek V4.1 Flash open-weight?

Yes. The weights are on [Hugging Face](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017) and ModelScope, and the model card says both the repository and the weights are under the MIT license. The files are large: the card lists the checkpoint at about 763B parameters in mixed BF16, F32, FP8 and INT8 formats. No chat template file is included; DeepSeek ships a Python reference encoder instead.

## What DeepSeek says about performance

The model card compares Flash-1017 with the earlier Flash. These are DeepSeek's own numbers, and we have not seen independent tests of them.

| Benchmark | Flash-1017 | Flash |
| --- | --- | --- |
| HLE (no tools) | 42.6 | 36.8 |
| Terminal-Bench 4.0 | 42.3 | 31.2 |
| DeepSWE v1.1 | 77.0 | 74.2 |
| ProgramBench | 29.1 | 20.3 |
| NL2Repo-Bench | 68.2 | 65.4 |

The card lists higher scores for Flash-1017 on every benchmark in its table, with the largest gains on Terminal-Bench 4.0 and ProgramBench. It also compares the model with outside models, including Opus 5. DeepSeek's table shows Flash-1017 ahead of Opus 5 on three of the benchmarks and behind on most of the others.

The release note also says the model was tuned with reinforcement learning from its own feedback for building whole projects, such as games and 3D scenes, and is optimized for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy.

## What we don't know yet

DeepSeek's pages don't state a release date for the base V4.1-Flash model, so we can't give one. The only dated release we can confirm is the October 17 update. We also have no independent benchmark results, so the comparison with other labs' models rests on DeepSeek's table.

The next date to watch is October 31, 2026 at 16:00 UTC, when DeepSeek says billing starts for its search and fetch tools at $0.75 per 1,000 calls.

## Frequently asked questions

### What is DeepSeek V4.1 Flash?

It is DeepSeek's current Flash model, a multimodal mixture-of-experts model that takes images and text and writes text. The latest version is DeepSeek-V4.1-Flash-1017, released on October 17, 2026.

### How much does DeepSeek V4.1 Flash cost?

As of October 29, 2026, the API charges $0.15 per 1M input tokens on a cache miss and $0.60 per 1M output tokens off-peak. Peak prices are double.

### Is DeepSeek V4.1 Flash open source?

The weights are public on Hugging Face and ModelScope, and the model card says the repository and weights are under the MIT license.

### What is the DeepSeek V4.1 Flash context window?

The context window is 1M tokens, with a maximum output of 384K tokens on the API.

## Sources

- [DeepSeek-V4.1-Flash-1017 release note](https://api-docs.deepseek.com/news/news261017), DeepSeek
- [Models & Pricing](https://api-docs.deepseek.com/quick_start/pricing), DeepSeek
- [DeepSeek-V4.1-Flash-1017 model card](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017), Hugging Face
- [DeepSeek-V4.1-Flash model card](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash), Hugging Face