model-release

HuggingModels announces the release of Qwen3.6-35B-A3B-Uncensored-Genesis-Hermes-V7-GGUF, a fine-tuned and quantized version of the Qwen model.

Meta releases Muse Code, a terminal coding agent, and Muse Spark 1.2, an upgraded coding-focused model with improved code generation, debugging, and long-horizon task handling.

Speculation that Gemini 3.5 Pro is coming soon, likely from Google.

Ling-3.0-flash MXFP4, a quantized model, has been released and runs locally on a single DGX Spark, achieving ~80 tok/s decoding and 2,500-3,500 tok/s long-input prefilling, enabling private on-device inference for coding, agents, and offline batch jobs.

Qwen developers held an AMA on Twitter, confirming an upcoming 27B model, sharing details about Qwen 3.8's 2.4T parameters and 95B active, and noting the community's influence on releases.

Intern-S2 Mobius is a Qwen3.5-35B derived model with an architectural difference claimed to improve throughput and reduce token consumption.

A user asks about Mach-1 Additive, a model reportedly achieving 95% of Qwen 3.6 35B's performance while being 10x smaller, questioning why it isn't more widely discussed.

Liquid AI released LFM2.5-2.6B, a new tiny model focused on agentic capabilities. The author is excited to test it for high-volume tasks like document summarization.

Ilya Sutskever's Safe Super Intelligence (SSI) is reportedly set to release its first model this month, according to Gavin Baker in a timestamped interview.

Ling-3.0-flash's weights are now downloadable under MIT (~128GB FP8), demonstrating ability to generate webpages across multiple design styles using pure code—CSS gradients, SVG paths, typography, and layout—without external images.

Alpamayo 2 Super, an open reasoning model for autonomous vehicles, is released on Hugging Face under the Linux Foundation's permissive OpenMDW-1.1 license, enabling commercial adaptation and redistribution.

A developer excitedly queues tests for Qwen's new 27B model, which Qwen promises brings a whole new level of capability.

InclusionAI released the Ling-3.0-flash model weights on Hugging Face with MIT license, including BF16 and official FP8 versions. The model uses a fine-grained MoE architecture with 512 experts and 8 active per token, totaling 127.5B params with 5.1B active.

inclusionAI released Ling-3.0-flash, a native hybrid reasoning model with 124B total/5.1B active parameters using a hybrid linear attention architecture (KDA+MLA) and sparse MoE. It matches or outperforms its 1T-class predecessor Ring-2.6-1T while being far more compute-efficient, with built-in agentic and long-context optimizations.

Discusses NousResearch's ongoing work on the Hermes agent, noting the 0.2 release in mid-March and questioning whether it can compete with end-to-end omni models like GPT-omni or PersonaPlex.

Ling-3.0-flash is a new model that the author tested and found capable of fixing hard software bugs that Qwen3.6-27b could not, with speed similar to DeepSeek V4 flash. Its release has been delayed to August 6.

Reddit post sharing Artificial Analysis scores for Qwen 3.8 Max, praising it as a cheap replacement for Opus 4.7.

SupraLabs releases Supra2-100M Base and Instruct models, a new small language model family with community-driven improvements, benchmarks, and a GGUF version.

A tweet highlights that major open-source models like Qwen3.8-Max may use linear attention, referencing Alibaba's announcement of upcoming open-weight releases for Qwen3.8-Max (2.4T parameters) and Qwen3.8-27B.

Meituan released LongCat-Flash-Lite-Sparse, a sparse-attention model supporting 1M-token context with only ~3B active parameters per token, achieving strong SWE-Bench scores under an MIT license.