DeepSeek has previewed V4 Flash and V4 Pro, mixture-of-experts models with 284B and 1.6T parameters featuring 1 million token context windows. The company claims the models nearly close the gap with frontier models on reasoning benchmarks and match GPT-5.4 on coding, while lagging 3-6 months behind on knowledge-based reasoning. Notably, V4 models are priced substantially below competing frontier models.
Models
DeepSeek previews new AI model that ‘closes the gap’ with frontier models
DeepSeek's V4 Flash and V4 Pro mixture-of-experts models claim parity with GPT-5.4 on coding and frontier reasoning benchmarks while underpricing competitors by a substantial margin.
Friday, April 24, 2026 12:00 PM UTC2 MIN READSOURCE: TechCrunchBY sys://pipeline
Tags
models
/// RELATED
Safety4d ago
Credit cards are vulnerable to brute force attacks
Payment gateways leak validation state through error codes, enabling attackers to brute-force the 4 missing card digits and bypass 3D Secure exemptions to steal funds despite PCI DSS masking rules.
Strategy4d ago
Artemis III aims for 'late 2027' for Earth orbit demonstration
NASA targets late 2027 for Artemis III Earth orbit demonstration of SpaceX and Blue Origin landers, setting up a 2028 lunar landing attempt with interoperability testing in between.