DeepSeek V4.1 Flash: one million tokens and open weights

DeepSeek V4.1 Flash is the update DeepSeek released on October 17, 2026 for its fast model: it reads up to one million tokens of context, writes up to 384 thousand tokens in a single response, and ships with open weights you can download. DeepSeek V4.1 Flash pricing is $0.60 USD per million output tokens off-peak, and double that at peak hours.

What changed in DeepSeek V4.1 Flash

The name deepseek-flash is now served by DeepSeek-V4.1-Flash-1017, and the way you call it does not change. The old names, deepseek-v4-flash and deepseek-v4-flash-vision-exp, still work, but those models are retired: requests are redirected to this version and billed at the Flash price.

DeepSeek describes it as a Pareto improvement in performance and cost: stronger on benchmarks, faster on tasks, and with shorter outputs.

What it does: context, output and price

• Reads: up to 1 million tokens of context, enough to hand it a whole project.
• Writes: up to 384 thousand tokens of maximum output.
• Charges: $0.60 per million output tokens off-peak and $1.20 at peak. Uncached input costs $0.15 off-peak and $0.30 at peak.
• Cache: reading from cache costs $0.003 per million tokens off-peak, under half a cent, and $0.006 at peak.
• Download: the open weights are on Hugging Face and ModelScope.

The hour matters: list price at peak hours

Peak hours run from 01:00 to 04:00 and from 06:00 to 10:00 UTC, Monday to Friday, not counting Chinese public holidays. Outside that window the price is half. If your agent runs long jobs, scheduling them off-peak is the most direct way to pay less.

How to apply it in your business?

Instead of rationing tokens, you can let an agent read a project's entire repository. To estimate your cost, take the output tokens your agent spends in a month, divide them by one million and multiply by $0.60 if you run off-peak. It is a reference calculation with list prices: it does not include input, cache or your real schedule.

If you want to compare with other long-context models, read the full breakdown of GPT-6.1 Sol with 500 thousand tokens.

Limits worth knowing

The official page does not state the license of the weights, so check it on Hugging Face before using them in a product. The benchmark results DeepSeek published come as an image, with no figures in the text, which is why this summary does not repeat them.

Frequently asked questions about DeepSeek V4.1 Flash

How much does DeepSeek V4.1 Flash cost?

$0.60 USD per million output tokens off-peak and $1.20 at peak, according to DeepSeek's pricing page.

How many tokens of context does it have?

One million tokens of context and up to 384 thousand tokens of maximum output.

Can I download DeepSeek V4.1 Flash?

Yes, the open weights are on Hugging Face and ModelScope. The license does not appear on the official announcement page.

What happens to deepseek-v4-flash?

It keeps working, but its requests are served by DeepSeek-V4.1-Flash-1017 and billed at the Flash price.

Conclusion

DeepSeek V4.1 Flash combines a one-million-token context, an output of up to 384 thousand tokens and a price that halves off-peak. Before migrating, check your usage hours and the license of the weights.