Fetching from the wire…
01
02
03
04
05
06
07
08
Models2026-08-30 · source-backed
Hy4 preview, released August 28: 770B total parameters, 49B active, over 1M token context, open-sourced and simultaneously on Tencent Cloud TokenHub and OpenRouter at $0.834 per million input tokens, $2.501 per million output, $0.042 per million cached. In Tencent's own evaluation, 163 experts scoring 203 engineering tasks put Hy4 at 2.99/4.00 against GLM-5.3 at 2.92 and Kimi K3 at 2.94. (Tencent) The distribution move is what I'd watch: Tencent bundled it free for two weeks into WorkBuddy and CodeBuddy, its office and coding products, instead of selling it as a standalone subscription. It also took #2 on Product Hunt's August 29 board with 202 votes, which is unusual for a raw model launch on a board built for consumer products.
Each link below shares sources, entities, or timing with this story.
Released August 28 with 78 layers, 77 of them MoE with 256 routed plus one shared expert and top-8 routing, plus a native 10B MTP layer for speculative decoding (GitHub). The attention stack uses Gated DeepSeek Sparse Attention with IndexCache for cross-layer sparse index reus...
A10 Networks made its AI Gateway generally available on August 14, pitched as a "centralized control plane for unified routing, cost management, and governance across every AI agent, application and large language model" (Help Net Security). SelectHub launched DataGrout the sa...
August 15's board leads with Inferock Bench at 154 upvotes, a local open-source proxy that sits between your app and OpenAI, Anthropic, Gemini or OpenRouter-shaped calls and emits a per-call billing-integrity receipt. It catches streams cut mid-sentence that were still billed,...
78 layers where layer one is dense FFN and the other 77 are MoE, each with 256 routed experts and 1 shared expert, top-8 routing per token, plus a native 10B MTP layer (0.7B activated) built in for speculative decoding. FP8 and base variants released together on August 28; the...
It showed up August 20 with 1,048,576 tokens of context, 131,072 max output, and text, image and video input. Stripe's Patrick Collison called it "very impressive," no lab has claimed it, and speculation splits between Z.ai's GLM family, Microsoft's MAI and Xiaomi's MiMo. The...
An open-weight Chinese frontier model is now a dropdown option in Microsoft's coding product. That happened before anyone finished characterizing what the model does. GitHub's changelog dated August 6 makes Kimi K3 generally available across Copilot Pro, Pro+, Max, Business an...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.