DeepSeek releases V4.1 Flash 1017 Max with 1M token context and reasoning ability

What was released

DeepSeek has released DeepSeek V4.1 Flash 1017 (Max), a reasoning-capable AI model with open model weights, according to a page on Artificial Analysis.

The model arrived in October 2026 and carries an MIT license, meaning anyone can use or modify the weights. DeepSeek said this version supports both text and image input, and it outputs text.

How it performs

Artificial Analysis gave the model 4 out of 4 units across Intelligence, Speed and Verbosity, while scoring it 2 out of 4 for Cost on its benchmark index.

The model holds a high intelligence ranking in its class, landing near the top of over 700 evaluated open-weight models. It also runs at about 225 output tokens per second, which the site describes as notably fast.

One notable trade-off is verbosity. Artificial Analysis recorded 210 million output tokens during its benchmark tests, well above the median figure for similar models.

Key specifications

The model uses a mixture-of-experts architecture with 552 billion total parameters but activates only 16 billion per token during inference. That design lets it handle large tasks without needing to run the full model.

It supports a context window of one million tokens, roughly equivalent to 1,500 A4 pages of standard text. That makes it suitable for processing very long documents or large codebases in one go.

Pricing data from Artificial Analysis lists input costs at $0.30 per million tokens and output costs at $1.20 per million tokens, with a 98 percent cache discount bringing effective prices down further.

The model does not currently have a publicly available API through Artificial Analysis.

What the benchmarks measure

Artificial Analysis evaluates models using its Intelligence Index, which combines ten separate tests including coding challenges, reasoning tasks, and knowledge checks. DeepSeek V4.1 Flash 1017 (Max) scored 45 on that index, compared with a median of 19 among peer models.

The testing covers a mix of general ability and domain-specific skills, including finance, engineering and strategy.