OPEN WEIGHTS

Strata runs the 125B Qwen3.8-Flash-Next on a 12 GB GPU with 64 GB of RAM at up to 94 tokens per second. Hardware needs, speeds by quant, the license catch, and when the $0.15 hosted API is the better call.

Kolibri is an Apache 2.0 English-German MoE with 78B total and 3B active parameters and a 1M context. Benchmarks, the vLLM command, and who should run it.

Anthropic's Frontier Red Team says GLM-5.3 builds working exploits at rates close to Claude Mythos Preview, and that its safeguards can be stripped in days. Here are the verified numbers, the community backlash, and the practical read for developers.

An independent project fine-tunes Qwen3.5 and Gemma 4 into 0.8B and 2B decision models that answer in 22-28 ms using Jev's request shape. The verified numbers, the run commands, and where the benchmark stops matching real work.

Tencent's Hy4 preview ships 770B total parameters with 49B active under Apache 2.0 - a 1M-context text MoE with DeepSeek-style sparse attention, posted Terminal-Bench 85.4 and DeepSWE 64.3, and an OpenRouter price of $0.834/$2.501. Verified against the model card and the live OpenRouter page on August 31, 2026.

DeepSeek V4.1 Flash replaced V4 Flash. Official API: $0.15 input, $0.60 output off-peak. OpenRouter hosts start near $0.05. MIT weights are on Hugging Face.

Qwen3.8-27B is a 27B dense Apache-2.0 model that scores 61.7 on SWE-bench Pro and 42.2 on DeepSWE 1.1 - ahead of Opus 4.6 Max on both - while running on consumer hardware. Benchmarks, hardware math, and an honest when-to-use-it guide.

GLM-5.3 routes compared, verified October 1, 2026: OpenRouter hosts list it from $0.12 per million input tokens, GLM-5.3-Flash costs $0.15 input and $0.50 output from Z.ai, batch drops to $0.45/$2.00, and the Coding Plan starts at $18 per month.

Cloudflare published the serving playbook behind Workers AI running Moonshot Kimi K2.6 and Zhipu GLM 5.2: FP8 KV caches double Kimi's resident context to 1.37M tokens, INT4 weights shrink GLM 5.2's checkpoint 40%, and a page-tagging integrity check protects the shared cache at under 1% overhead. The numbers show what actually matters when open frontier models run on GPU fleets.

Alibaba released Qwen 3.8 Max on August 3, 2026 - a 2.4T-parameter MoE with 95B active per token, a 1M context window, and $2/$6 per million tokens on QwenCloud. It leads PaperBench at 93.0, and the weights open next week.

Kimi K3 open weights need roughly 1.5TB of VRAM, which does not fit on a B200 node. That forces a real hardware decision: B300, two B200 nodes, or AMD's MI355X. Here is the head-to-head with verified specs, the Wafer benchmark, and what it costs per token.

Dario Amodei published Anthropic's stance on open-weights models this week - no total ban, but support for chip export controls, distillation crackdowns, and mandatory safety testing. HN responded with 800+ comments calling it regulatory capture. Here is what the CEO said, what the thread argued, and why the debate matters for every developer deploying AI.

Moonshot AI's Kimi Linear paper introduces KDA, a hybrid linear attention that beats full attention at all scales - 75% less KV cache, 6x decoding at 1M context, and open-source checkpoints.

Moonshot AI released the full Kimi K3 weights on HuggingFace today - 2.8T parameters, 1M context, native MXFP4 quantization, ~1.63TB download. The HN community reaction, what the license really says, and why this matters for the open-weights AI market.

Where to access Kimi K3: Moonshot's API at $3 input and $15 output per million tokens, OpenRouter hosts, inference clouds, and open weights.

Moonshot AI releases Kimi K3 with 2.8 trillion parameters, 1M context window, and Delta Attention architecture. Here's what developers need to know about pricing, performance, and where it fits in the frontier model landscape.

Martin Alderson's argument for why open-weights models like GLM 5.2 will compress frontier lab margins is sparking debate on HN. Here is what the thesis actually says, where HN agrees and disagrees, and why it matters for developers choosing models.

Unsloth's dynamic quantization makes GLM-5.2 runnable on a 256GB Mac or a 24GB GPU with CPU offloading. Here is the hardware math, the quantization tradeoffs, and what the HN community learned from actually running it.

GLM-5.2 is an MIT-licensed open-weights coding model. Cheapest routes: OpenRouter hosts from about $0.36 per million input tokens, Z.ai's API, or self-host.

New benchmark data shows GPT-5.5 hallucinates 86% of the time when it does not know the answer - versus 28% for the open-weights GLM-5.2. The numbers challenge the assumption that bigger models equal more reliable output.