DeepSeek-V4-Pro launches on CoreWeave with 1.6T params, 1M context
wandb · x · 2026-09-02
DeepSeek-V4-Pro-0813 is live on CoreWeave Serverless Inference. Featuring 1.6T parameters and a 1M context window, it is built for long-horizon agent work. Cache reads are priced at $0.044/M, optimizing for the context agents resend at every step.
More from Infra
- B300 Supports 6x More Concurrent Agents Than H200: Benchmark — ryanshrout · 2026-09-02
- AMD's Primus tuning agent predicts config performance to save thousands of GPU-hours — PyTorch · 2026-09-02
- NeoCloud Market Heats Up: Vendors Compete Beyond Raw GPU Access — DavidLinthicum · 2026-09-02
- Kernel rewrites Go proxy for agents, speeding up wakeups by 92% — ycombinator · 2026-09-02
- Benchmark Scores Don't Matter; Open Local Models Are the Future — johnseach · 2026-09-02
- Bittensor SN44 to evolve from vision to world simulation — richdotca · 2026-09-02