Unsloth shipped an update adding local support for Recursive's Flint 40B coding-agent MoE, OpenMOSS's 5.1B MOSS-Image generator, and NVIDIA's Nemotron 3.5 Bolt E4B reasoning fine-tune. The release also adds an OpenAI-compatible fine-tuning API, REAP MoE expert pruning, SSD-offloaded MLX MoE inference for oversized Mac models, and a chat-only Unsloth Desktop install. For practitioners running open models locally, it expands the single-tool path to fine-tune, quantize, and serve several of the newest open-weight releases.
Unsloth shipped an update enabling local runs of NVIDIA's new Nemotron 3.5 models and Meta's new Muse Spark Open, plus training support for text-diffusion LLMs, DeepSpeed training, and improved prompt caching for coding agents. The update matters to practitioners running frontier-scale open models on local hardware, with quantized memory footprints detailed for each model.
Unsloth shipped an update enabling local runs of NVIDIA's new Nemotron 3.5 models and Meta's new Muse Spark Open, plus training support for text-diffusion LLMs, DeepSpeed training, and improved prompt caching for coding agents. The update matters to practitioners running frontier-scale open models on local hardware, with quantized memory footprints detailed for each model.
Ollama released v0.40.10 adding support for NVIDIA's Nemotron 3.5 Bolt E4B, a Gemma 4-based vision-language model for fast local conversations, image understanding, and tool use. The release makes the model runnable locally via `ollama run nemotron-3.5-bolt:e4b`. It gives developers an option for local multimodal inference with speculative decoding.
Ollama released v0.40.10 adding support for NVIDIA's Nemotron 3.5 Bolt E4B, a Gemma 4-based vision-language model for fast local conversations, image understanding, and tool use. The release makes the model runnable locally via `ollama run nemotron-3.5-bolt:e4b`. It gives developers an option for local multimodal inference with speculative decoding.
OpenMOSS has released MOSS-Image, an image generation model published on Hugging Face alongside a base model variant. Like all OpenMOSS releases, both are Apache-licensed, making them fully open for commercial use. Community members are already asking how it compares to Qwen Image 2.1 and Flux Klein 3.
OpenMOSS has released MOSS-Image, an image generation model published on Hugging Face alongside a base model variant. Like all OpenMOSS releases, both are Apache-licensed, making them fully open for commercial use. Community members are already asking how it compares to Qwen Image 2.1 and Flux Klein 3.
Recursive, the startup led by Richard Socher and cofounded by Jeff Clune, released Flint, a free, open-weight coding model that runs on a laptop. The company says its automated research system, the Eureka Machine, contributed to nearly every step of development, letting a small team build the model from scratch in under four months. The launch is an early public test of recursive self-improvement — AI building better AI — a controversial idea facing both extinction warnings and proposed Congressional bans.
Recursive, the startup led by Richard Socher and cofounded by Jeff Clune, released Flint, a free, open-weight coding model that runs on a laptop. The company says its automated research system, the Eureka Machine, contributed to nearly every step of development, letting a small team build the model from scratch in under four months. The launch is an early public test of recursive self-improvement — AI building better AI — a controversial idea facing both extinction warnings and proposed Congressional bans.
NVIDIA released the Nemotron 3.5 family as open weights: Super VL 120B-A12B, a multimodal model built for long-running agents, plus Bolt 74B-A9B (24GB) and Bolt E4B (4GB) edge models. ZDNET reports Super VL benchmarks 5x faster than the next best open agentic model per Artificial Analysis, with InclusionAI's Ling 3.0 Flash and DeepSeek's V4.1 Flash behind it. The releases land amid a Western surge in open-weight models from Mistral, Reflection AI, and others challenging Chinese labs.
NVIDIA released the Nemotron 3.5 family as open weights: Super VL 120B-A12B, a multimodal model built for long-running agents, plus Bolt 74B-A9B (24GB) and Bolt E4B (4GB) edge models. ZDNET reports Super VL benchmarks 5x faster than the next best open agentic model per Artificial Analysis, with InclusionAI's Ling 3.0 Flash and DeepSeek's V4.1 Flash behind it. The releases land amid a Western surge in open-weight models from Mistral, Reflection AI, and others challenging Chinese labs.