Loading...
DeepSeek's new model brings frontier AI performance at ultra-low cost
The new model comes with 552 billion parameters

DeepSeek's new model brings frontier AI performance at ultra-low cost

Sep 11, 2026
10:42 am

What's the story

DeepSeek has unveiled its latest artificial intelligence (AI) model, DeepSeek-V4.1-Flash, which is said to outperform some of the industry's leading models like OpenAI's GPT-5.6 Sol and Anthropic's Claude Opus 5. The new model comes with a 552-billion-parameter mixture-of-experts backbone and a native vision capability. It also features a one-million-token context window and an architecture designed for cost-effective reading of large contexts repeatedly.

Pricing details

DeepSeek's competitive pricing strategy

DeepSeek has set a competitive pricing strategy for its new model.

During off-peak hours, the company charges $0.003 per million input tokens on a cache hit, $0.15 per million on a cache miss, and $0.60 per million output tokens. However, peak rates are double those figures.

This pricing structure is especially beneficial for developers testing the model for coding agents and other long-running workflows, as they may reread more cached context than fresh context over time.

Performance comparison

Comparing costs with other leading AI models

When compared to other leading AI models, DeepSeek-V4.1-Flash's off-peak cached-input rate is surprisingly low.

OpenAI's GPT-5.6 Sol costs $4 per million regular input tokens, $0.40 for cached input, and $20 for output while Anthropic charges $5 for standard input, $0.50 for Claude Opus 5 cache hits, and $25 for output.

Moonshot AI's Kimi K3 comes at a cost of $3 for cache-miss input, $0.30 for cache-hit input, and $15 for output tokens.

ADVERTISEMENT

Cost efficiency

DeepSeek outperforms competitors in cost efficiency for coding agents

The cost efficiency of DeepSeek-V4.1-Flash is particularly notable for coding agents that hit a cache across multiple requests.

For instance, an agent with a 500,000-token reusable prefix hitting that cache across 100 requests would incur a cost of about $0.15 on DeepSeek-V4.1-Flash off-peak.

This is significantly lower than the costs incurred by other models such as Kimi K3 ($15), GPT-5.6 Sol ($20), and Claude Opus 5 ($25) at their respective published cache-read rates.

ADVERTISEMENT