Ling-3.1-flash is now open source with 560 B total parameters, about 25 B active per token and a 1 M-token context window. It can turn rough briefs into pages, mobile UIs, and interactive prototypes. Model weights and documentation are hosted on Hugging Face and ModelScope.
Key points
Open source release: weights and docs on Hugging Face and ModelScope
Specs: 560 B total parameters, ~25 B active per token, 1 M context length
Sources
- Original post
- inclusionAI/Ling-3.1-flash - Linked page
- Linked resource - Linked in the post
Sorin Trading Copilot upgraded, Hyperliquid integration
Sorin Trading Copilot is upgraded and now integrates with Hyperliquid. Users select AI Analysis, describe a trade, and Sorin provides reasoning, settings, and fills the order card after confirmation.
Key points
Upgrade: Sorin's Trading Copilot is upgraded.
Integration: The upgraded Copilot includes Hyperliquid integration.
Sources
Huihui-Qwen3.8-Flash-Next-abliterated-v2-GGUF new release announced
Huihui-Qwen3.8-Flash-Next-abliterated-v2-GGUF is an uncensored version of Qwen3.8-Flash-Next created with abliteration. The release notes that ablation emphasis has been adjusted and no experts were ablated. The unsloth series now supports the latest versions of Strata and llama.cpp.
Key points
Release: New GGUF V2 (unsloth series, Strata) for Huihui-Qwen3.8-Flash-Next-abliterated-v2-GGUF.
Change: Ablation emphasis adjusted; none of the experts have been ablated.
Sources
MiniCPM-V 4.7 OpenVINO support on Intel CPUs GPUs NPUs
MiniCPM-V 4.7 now has day-0 OpenVINO support for Intel CPUs, GPUs and NPUs, enabling multimodal AI workloads like document understanding, OCR and video analysis. The announcement thanks Intel developers for the support.
Key points
Day-0 OpenVINO support for MiniCPM-V 4.7.
Multimodal AI workloads for document understanding, OCR, and video analysis run on Intel CPUs, GPUs, and NPUs.
Sources
AutoEnvScaling introduces self-building Harbor tasks
AutoEnvScaling turns Data Flywheel into a terminal task where a proposer agent builds complete Harbor tasks and sandbox. This enables self-play RL for terminal agents, allowing one model to improve as both builder and solver without human-in-the-loop.
Key points
Proposer agent builds complete Harbor tasks and sandbox.
Enables self-play RL for terminal agents, one model improves as both builder and solver, without human-in-the-loop.
