From Startup Fortune · · 1m 10s
DeepSeek Ships Cheaper AI Model
DeepSeek quietly releases V4.1-Flash-1017 model beating rivals on agent tasks.
In brief
DeepSeek quietly releases V4.1-Flash-1017 model beating rivals on agent tasks. DeepSeek's V4.1-Flash-1017 model improves software engineering benchmarks at lower cost than Western rivals. Originally reported by Startup Fortune.
DeepSeek Releases Model
DeepSeek dropped V4.1-Flash-1017 on Hugging Face two days ago without a launch event.
Model Improves Benchmark Scores
The update pushed DeepSWE scores from 74.2 to 77.0, beating all 11 benchmarks by 2.0 to 11.1 points.
Model Architecture Details
V4.1-Flash-1017 uses a causal encoder-decoder design with 552 billion parameters.
Cost Efficiency Focus
Only 8 billion parameters activate while reading prompts, and 16 billion while writing responses.
Search API Opens
DeepSeek opened its proprietary Search API to the public for the first time.
Permanent API Discount
DeepSeek made its 75 percent V4-Pro API discount permanent after the promotion period.
Price Comparison
Peak-hour API pricing sits at $0.30 per million input tokens and $1.20 per million output tokens.
Cost Advantage
DeepSeek's pricing is half the cost of Anthropic's Claude Opus or OpenAI's frontier-tier models.
The headline
DeepSeek's V4.1-Flash-1017 model improves software engineering benchmarks at lower cost than Western rivals.
