July 13, 2026

Agentic Frameworks, Tooling, Skills

GitHub Trending Python: MCP Knowledge-Graph & Transcript Router, https://github.com/trending/python?since=daily — Trending Python utility exposing an MCP server + real-time WebSocket transcript handler for Google Meet/Teams/Zoom; replaces ad-hoc scraping hooks with a self-hostable agentic skill for deterministic telemetry/data routing into local vector/graph stores. [Source: https://github.com/trending/python?since=daily]

Notable Research

Long-Horizon-Terminal-Bench, https://huggingface.co/papers — Tencent Hunyuan’s dense-reward grading harness for long-horizon terminal agents; provides a structured drift-tracking and horizon-collapse baseline for rover autonomy loops and multi-step tool-use evaluation without external simulators. [Source: https://huggingface.co/papers] KronQ: LLM Quantization via Kronecker-Factored Hessian, https://huggingface.co/papers — Quantization method leveraging Kronecker-factored Hessian approximations; improves weight-mapping fidelity during aggressive quantization, offering a path to lower-bit local inference without the accuracy collapse seen in standard GGUF variants. [Source: https://huggingface.co/papers] Self-Guided Test-Time Training for Long-Context LLMs, https://huggingface.co/papers — Addresses KV-cache degradation in extended context; introduces self-guided training mechanisms that stabilize attention decay, directly applicable to extending your num_ctx=32k base without RAG overhead. [Source: https://huggingface.co/papers]

Frontier Lab Updates

Anthropic/Fable 5 Access Policy Shift, https://simonwillison.net/ — Anthropic now restricts Fable 5 to 50% of weekly usage limits on Claude Max plans, forcing agentic routing architectures to implement hard fallbacks to other model tiers rather than relying on Fable as a persistent execution backbone. [Source: https://simonwillison.net/] OpenAI/GPT-5.6 Sol Efficiency Rollout, https://simonwillison.net/ — GPT-5.6 Sol now deploys broader efficiency optimizations and retains background delegation to GPT-5.5 for GPT-Live; reduces latency budgets for complex tool execution while maintaining conversational flow, shifting local routing decision logic toward structured offloading boundaries. [Source: https://simonwillison.net/]

Models to Download & Try

unsloth/Qwen3.6-27B-NVFP4, https://huggingface.co/unsloth/Qwen3.6-27B-NVFP4 — 21B parameters; ~12GB Q4_K_M footprint leaves ~20GB for KV context; NVIDIA’s FP4 quantization variant of the 35B class stack; reduces VRAM pressure for agentic tool-calling loops while preserving instruction-following fidelity. [Source: https://huggingface.co/models?sort=trending] robbyant/lingbot-video-moe-30b-a3b, https://huggingface.co/robbyant/lingbot-video-moe-30b-a3b — 30B MoE (3B active); ~16–18GB Q4_K_M footprint fits in 32GB with ~14GB context headroom; vision-to-video generation specialist; plugs into the vision-scraper pipeline for high-fidelity aerospace schematic reconstruction and rover telemetry simulation without external API dependencies. [Source: https://huggingface.co/models?sort=trending] GnLOLot/MiniCPM5-1B-Claude-Opus-Fable5-Thinking-GGUF, https://huggingface.co/GnLOLot/MiniCPM5-1B-Claude-Opus-Fable5-Thinking-GGUF — 1B parameters; ~0.6GB footprint leaves >31GB for context/reasoning; ultra-lightweight routing/thinking layer positioned to filter complex sub-tasks away from the 35B backbone on constrained memory. [Source: https://huggingface.co/models?sort=trending]

Skipped as Already Covered

• tencent/Hy3 (295B/299B MoE; exceeds 32GB VRAM limits)
• zai-org/GLM-5.2 (753B cloud-scale; covered 7/6/7/8/7/10)
• empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF (ultra-long context; covered 7/4/7/5/7/8/7/10/7/11/7/12)
• InternScience/Agents-A1 (35B agentic orchestration; covered 7/5/7/8/7/10/7/11/7/12)
• HauhauCS/Qwen3.6-35B-A3B-Uncensored (repack/uncensored variant; covered 7/5/7/8/7/12)
• Meta/Muse Spark 1.1 API / OpenAI GPT-5.5 delegation / Anthropic Claude tool-schema drift (frontier routing updates; covered 7/9/7/10/7/11)