Claude Code・Codex・Piを比べる検証で、UCバークレーの博士課程の研究者が、モデルを取り巻く仕組みが生む隠れた負荷(harness tax)を調べた(arena)。周囲の実行環境(ハーネス)の作りが、コードを書く・画面を操作する実作業の出来を左右する仕組みがある。この作り込みの差が成果を左右する要素になっていく可能性がある。
ALERT 🚨🚨OpenAI's latest GPT6.1 Sol Ultrafast is NOT running on Cerebras but is instead running at a low batch size on NVIDIA GPUs.
What does this say about Cerebras? Will Cerebras be serving GPT6.1 Sol Ultrafast in the future?
GENERAL COMPUTE TO DEPLOY CEREBRAS CHIPS FOR AI INFERENCE
General Compute is expanding its AI infrastructure with a new Cerebras deployment alongside NVIDIA GPUs, aimed at improving inference speed and economics.
The company says it will use a disaggregated setup, with GPUs handling prefill and Cerebras systems handling inference.
The idea is to use different types of compute for the parts of the workload where each performs best.
General Compute recently raised $400M in debt to help finance alternative AI hardware deployments, including Cerebras systems.
General Compute says its first Cerebras-powered tokens are expected to go live in Q1 2027.
The Information: DeepSeek’s ARR has reached $1 billion.
DeepSeek allocates 70% of its compute capacity to training and 30% to inference.
DeepSeek is using NVIDIA gaming GPUs to run inference on smaller models, freeing up compute resources for training to help ease the shortage.