October 2026: Inference, agent tooling, and deployment costs
The Forward Pass research desk Data report
Editorial archive
Large 4’s coding pitch: equal scores, fewer tokens
The Forward Pass research desk Deep dive
Agent Memory Needs a Reader
Anthropic’s memory experiments show that retaining experience is only part of the system: models also need to retrieve it, test it, and keep it useful.
The Forward Pass research desk Deep dive
Voxtral 3 makes the reasoning model a pricing decision
The Forward Pass research desk Deep dive
The 530 GB decision behind Muse Spark Open
The Forward Pass research desk Deep dive
Strata’s 125B Local Runtime Is a Hardware-Sharing Story
Strata reports consumer-GPU inference for Qwen3.8 Flash Next by splitting expert execution across GPU and CPU, with speculative decoding adding a claimed 1.6 to 1.8× speedup.
The Forward Pass research desk Deep dive
Pi 1.0: a small agent harness grows Codemode, MCP and virtual models without getting big
Earendil's Pi hits 1.0 with Codemode, native MCP, virtual-model routing, deferred tool loading and Anthropic cache warming, plus an experimental Pi Durable package for long-running agents.
The Forward Pass research desk Deep dive
Cloudflare's Clef replaces text generation with schema scoring for agent decisions
Cloudflare has released Clef and Clef-flash, open-weight decision models on Workers AI that return structured classifications with probabilities, and it reports latency well below Jev's.
The Forward Pass research desk Deep dive
You’re all caught up