2026-10-17

3 source-first briefs published in this edition.

Technical editorial illustration of a multimodal input stream entering a radial mixture-of-experts routing network, with highlighted expert paths

Model release17 Oct 2026

Meta releases 816B Muse Spark Open weights for multimodal coding agents

Meta's 816B-total, 42B-active mixture-of-experts checkpoint accepts text, images, audio and video, but its gated weights and custom license require careful deployment review.

Primary sourceMeta Superintelligence Labs official Muse Spark Open model card and Muse Model LicenseSource published 2026-10-15

Read evidence brief →

Technical editorial illustration of three BF16, FP8 and NVFP4 model weight blocks aligned on a computational grid

AI-generated editorial illustration

Model release17 Oct 2026

Reflection publishes Beam 501B weights under Apache 2.0 after preview

Beam moves from an early-October announcement to downloadable 501B-total, 23B-active model weights in three numerical formats, plus an optional speculative-decoding drafter.

Primary sourceReflection official 16 October Beam availability announcementSource published 2026-10-16

Read evidence brief →

Technical editorial illustration showing an input block, a recurrent layer loop with one to four passes and an output block in a small reasoning model

AI-generated editorial illustration

Model release17 Oct 2026

NVIDIA releases Dory 4B open weights with adjustable latent-recurrent compute

The experimental 4B-parameter Dory model reuses a recurrent transformer block at inference time, offering adjustable loop counts rather than a larger checkpoint for each compute budget.

Primary sourceNVIDIA official Nemotron Labs Dory1.0-4B model cardSource published 2026-10-16

Read evidence brief →