Silicon
Cerebras: Faster Tokens Please
SemiAnalysis published a technical write-up on Cerebras wafer-scale inference, the Cerebras and OpenAI partnership, and the economics of fast token generation. The piece is dated 2026-05-13.
Source: SemiAnalysis
Evaluating Geekbench 6
Chester Lam at Chips and Cheese posted an evaluation of Geekbench 6, examining its instruction composition, branch prediction patterns, and memory access behavior across modern CPU microarchitectures including Intel Lion Cove and AMD Zen 5. Published 2026-05-07.
Source: Chips and Cheese
Weights
How open model ecosystems compound
Nathan Lambert posted a piece on Interconnects arguing that China’s open-first AI release pattern produces cost advantages by letting labs learn from peers and avoid duplicating research effort. Dated 2026-05-12.
Source: Interconnects
Runtime
vLLM v0.21.0rc2
vLLM tagged a release candidate v0.21.0rc2 on 2026-05-13. The noted change installs the nvidia-cutlass-dsl[cu13] extra on CUDA 13 platforms, continuing the v0.21.0rc1 work to package bundled DeepGEMM as a per-Python _C module so the wheel imports correctly.
Source: vLLM GitHub Releases
vLLM v0.20.2
vLLM v0.20.2 shipped on 2026-05-10 as a patch on the v0.20 stable line. It precedes the in-progress v0.21.0 release candidates.
Source: vLLM GitHub Releases
SGLang v0.5.11
SGLang published v0.5.11 on 2026-05-05. The release notes reference v0.5.10.post1 as the prior comparison point.
Source: SGLang GitHub Releases
Governance
Listening, Learning, and Building Together at OSI
Nick Vidal posted a one-month update from the Open Source Initiative on 2026-05-13, framed around a note from new executive director Duane O’Brien (now in his fourth week). The note outlines OSI’s planned presence at UN Open Source Week on 22-26 June 2026 and continues the organization’s focus on the Open Source AI Definition.
Source: Open Source Initiative
Import AI 456: RSI and economic growth; radical optionality for AI regulation; and a neural computer
Jack Clark’s Import AI 456 (2026-05-11) covers recursive self-improvement as a driver of economic growth, an argument for “radical optionality” as an AI-regulation approach, and a write-up of a neural-computer research direction.
Source: Import AI