RedditQwen 3.5 Small Series Natively Multimodal 262K Context Apache 2.0Reddit·high signalXBlueskyLinkedInCopy linkAlibaba releases Qwen3.5-0.8B to 9B natively multimodal 262K context Apache 2.0 with Gated DeltaNet architecture, 9B beats Qwen3-30BSourceSource pageReddit↳ Follow the threadStack layer / ContrastSwapping the N-gram layer of an IQ4 Qwen 3.8 Next quant to Q8 costs essentially no decode speedr/LocalLLaMAPolicy dependency / Stack layerTyped Provenance Guardrails Block All 19 Unsafe Releases From a Persistent Agent's Autobiographical MemoryarXiv 2609.02127Stack layer / Threat patternA web-browsing agent leaks recoverable signal about records it silently loaded, with no jailbreak and no disclosurearXiv 2609.01663Stack layer / Threat patternGPT-6-Astra propagated into four OpenAI-ecosystem SDKs within roughly 24 hours of the model landingGitHubPolicy dependency / Stack layerDRACO Redistributes One Rubric Score Across Agent Steps and Beats Ground-Truth-Reward GRPO by 5.3 Points on AppWorldarXiv 2609.04094Stack layer / ContrastOpenClaw 2.0 Ships With 16,000 Merged PRs and Detects Your Existing ChatGPT and Claude Subscriptions Instead of Asking for API KeysInfoQ (corroborated by Dataconomy)Stack layer / ContrastCodex shipped two hotfixes in 26 hours purely to carry GPT-6 Astra into the 0.153 lineGitHubStack layer / ContrastGLM 5.3 Flash wrote a working Fabric-API Minecraft mod locally in 7.6M output tokens over about 9 hoursr/LocalLLaMA