Breaking

DeepSeek Releases V4.1 Flash as a Lower-Cost Challenger to Astra and Fable

September 14, 2026 at 13:49 EDT

Foundation Models
AI Agents
Open Source

DeepSeek released DeepSeek-V4.1-Flash on September 10, turning its low-cost Flash line into a serious price-performance challenger for agentic and coding workloads. The new model is not presented as a clean overall leader over GPT-6 Astra or Claude Fable 5.1, but its published token prices are dramatically lower while several agent benchmarks put it in the same conversation.

Sep 10 · DeepSeek

DeepSeek V4.1 Flash pushes frontier-class agents below $1 per million output tokens

The model pairs a 1M-token context window, native vision and MIT-licensed weights with an off-peak output rate of $0.60 per million tokens. Its benchmark profile is competitive on agent work, though not an across-the-board lead over GPT-6 Astra or Claude Fable 5.1.

$0.60

Off-peak output per 1M tokens

Context window

74.2

DeepSWE v1.1 resolved score

552B

Backbone parameters

Output price gap

First-party standard output pricing per 1M tokens: DeepSeek off peak versus GPT-6 Astra and Claude Fable 5.1.

$0.60

DeepSeek

$50

GPT-6 Astra

$50

Claude Fable 5.1

Takeaway: DeepSeek's off-peak output token price is about 83x lower than either rival's standard rate.

Why developers care

Low token cost changes which coding, data-cleaning and agent tasks are worth automating, especially when repeated attempts are acceptable.

What still needs testing

Early trials show strong cost results but mixed reliability, with harder benchmarks and production routing decisions still requiring careful evaluation.

The strategic shift

Open weights→Lower cache cost→More agent attempts→Pressure on frontier pricing

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.