OPEN WEIGHTS

Mistral Large 4 weights are on Hugging Face under a modified MIT license. What the license blocks, the vLLM command, and when self-hosting beats the API.

DeepSeek V4.1-Flash-1017 jumps 11 points on Terminal-Bench 4.0 at the same deepseek-flash price. Benchmarks vs Opus 5, K3, GLM-5.3 and when to switch.

Meta released Muse Spark Open weights on October 15, 2026: 816B total, 42B active, 1M context. Run recipes, hardware floor, license catch, and where it trails.

DeepSeek 4.1 Flash as a daily coding model: what a session costs, how peak hours double the bill, why caching matters, and where the subsidy argument holds.

llama.cpp now caches hot MoE experts on the GPU. In the PR author's RTX 5090 run, --cpu-moe plus --moe-cache-mib decoded 2.2x faster than --fit. When to use each flag.

Kolibri 1 runs locally from 28.6 GB (Q2_K) to 83.1 GB (Q8_0). Which quant fits your RAM or Mac, measured speeds, and why Ollama cannot load it yet.

Step 5 Preview is StepFun's 600B MoE with 27B active and a 1M window. What it scores, what a session costs, and what to check before the October 15 weights.

Mistral Large 4 is a 1T-parameter multimodal MoE with 52B active, in public preview at $0.68/$2.09 per million tokens, with open weights due end of October.

Run Qwen3.8-Flash-Next locally with Strata on a 12 GB GPU: hardware needs, speeds by quant, the license catch, and when the 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 GLM-5.3 cyber report: exploit numbers against Claude Mythos Preview, how its safeguards were stripped, the backlash, and the read for developers.

Jeff fine-tunes Qwen3.5 and Gemma 4 into small Jev-compatible decision models on one GPU. The numbers, run commands, and where the benchmark misses real work.

Tencent Hy4 preview is a 770B open MoE with 49B active under Apache 2.0. Architecture, posted agentic benchmarks, OpenRouter pricing and self-hosting needs.

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 beats Opus 4.6 Max on SWE-bench Pro (61.7 vs 53.4). Benchmarks, hardware, cost and when to run it.

GLM-5.3 has no free hosted tier. Cheapest on October 9: Wafer via OpenRouter at $0.049/$3.39, Flash from $0.075/$0.25. Z.ai's promos ended October 7.

Cloudflare runs Kimi K2.6 and GLM 5.2 on Workers AI with FP8 KV caches, INT4 weights and a KV cache integrity check. What each technique buys on GPU fleets.

Qwen 3.8 Max is a 2.4T-parameter MoE with 95B active, a 1M context window and $2/$6 per million tokens on QwenCloud, with open weights promised next week.

AMD MI355X vs NVIDIA B200 vs B300 for serving Kimi K3's about 1.5TB of open weights: verified specs, the Wafer benchmark, and cost per token for each.

Dario Amodei on open-weights models: export controls, distillation crackdowns and mandatory safety testing, the HN pushback, and what it means for developers.