DeepSeek says R2 cuts inference cost by 32% on commodity GPU clusters

Full Summary

DeepSeek published R2 with open weights and accompanying technical notes focused on reasoning performance and inference efficiency. Early developer response centered on whether the release can pressure the closed-model leaders on cost and transparency at the same time.

Coverage Tags

Related Companies

Full Summary

[P] Fused MoE Dispatch in Pure Triton: Beating CUDA-Optimized Megablocks at Inference Batch Sizes I built a fused MoE dispatch kernel in pure Triton that handles the full forward pass for Mixture-of-Experts models. Model launches reshape the race because they force rivals to answer on capability, distribution, and rollout speed.

Coverage Tags

Related Companies

Full Summary

Show HN: ACE – A dynamic benchmark measuring the cost to break AI agents We built Adversarial Cost to Exploit (ACE), a benchmark that measures the token expenditure an autonomous adversary must invest to breach an LLM agent. This matters because it changes how the market reads current momentum, execution quality, or adoption potential.

Coverage Tags

Related Companies

Full Summary

Show HN: ACE – A dynamic benchmark measuring the cost to break AI agents We built Adversarial Cost to Exploit (ACE), a benchmark that measures the token expenditure an autonomous adversary must invest to breach an LLM agent. This matters because it changes how the market reads current momentum, execution quality, or adoption potential.

Coverage Tags

Related Companies

Full Summary

[D] The memory chip market lost tens of billions over a paper this community would have understood in 10 minutes TurboQuant was teased recently and tens of billions gone from memory chip market in 48 hours but anyone in this community who read the paper would have seen the problem with the panic immediately. Policy stories matter because compliance friction can slow adoption even when model quality keeps improving.

Coverage Tags

Related Companies

Full Summary

[P] Fused MoE Dispatch in Pure Triton: Beating CUDA-Optimized Megablocks at Inference Batch Sizes I built a fused MoE dispatch kernel in pure Triton that handles the full forward pass for Mixture-of-Experts models. Model launches reshape the race because they force rivals to answer on capability, distribution, and rollout speed.

Coverage Tags

Related Companies