Open-source reasoning model with frontier-grade coding at a fraction of the cost.
DeepSeek V3 is the pricing shock of 2026 — frontier-adjacent scores on MMLU-Pro and HumanEval at 5% of the cost of proprietary flagships. Open-weights, 128K context, strong at coding and math. The default choice for cost-sensitive production.
Get updates on new model releases, benchmarks, pricing changes, and AI research.