Release 5.17.0
Release v5.17.0 New Model additions HYV4 Hy4-Preview is a 780B-parameter mixture-of-experts language model that activates 49B parameters per token. Each MoE layer holds 256 routed experts plus one always-active shared expert and routes every token to 8 of them. The context window is 1M tokens. The architecture combines four features: Multi-head Latent Attention (MLA) compresses keys and values into a low-rank latent ( kv_lora_rank ) that kv_b_proj expands back to one key/value per query head. De
Read original source
Source record
Research analysis
What changed
Source account: Release v5.17.0 New Model additions HYV4 Hy4-Preview is a 780B-parameter mixture-of-experts language model that activates 49B parameters per token. Each MoE layer holds 256 routed experts plus one always-active shared expert and routes every token to 8 of them. The context window is 1M tokens. The architecture combines four features: Multi-head Latent Attention (MLA) compresses keys and values into a low-rank latent ( kv_lora_rank ) that kv_b_proj expands back to one key/value per query head. De
Version & cost evidence
Recent records to compare: “[分享创造] 从地心到太空的旅程 - 一个纯展示的小网站” (V2EX, Nov 5, 2026); “Nemotron 3.5 Transcribe” (Product Hunt, Nov 5, 2026). Shared themes are navigation hints; they have not been verified as the same need or independent corroboration.
Who is affected
Identify the existing workflow this change affects before recommending an upgrade. Launch attention does not establish improvement; an API, pricing or availability change may instead create migration work.
Missing comparison points
Missing checks: old versus new behavior, affected plans and regions, availability date, compatibility, rollback path and cost under the same workload.
Migration check
Save the release or policy notice, run the same task on the old and new option, and compare outputs and the full bill. For retirement notices, verify the deadline before scheduling migration.