Open Weights Forge

Local Fine-Tuning Tools

• 🔥 Unsloth Studio: Local-first web UI for the entire LLM lifecycle including search, download, chat, fine-tune, and...

Unsloth Studio is a local-first web UI for the full LLM lifecycle—search, download, chat, fine-tune, and export—right from your browser on your...

GLM-5 (Reasoning) surges to the highest-ranked open weights model on leaderboards with an Intelligence Index score of 50—that's among 195 open weights models out of 298 total evaluated.

• Forge enables pre-training, post-training (like fine-tuning), and RL on proprietary datasets for enterprise needs.
• Early users include ASML,...

New Open-Weight Releases

• 🔥 Mistral Leanstral: Mistral launched LeanStral, a family of compressed models from Mistral Large and Small using...

Mistral Forge lets enterprises build frontier-grade AI on proprietary data, covering what it does, why it matters, and comparisons to alternatives.

Mistral launches Leanstral and Small 4 models under Apache 2.0, optimized for open deployment in enterprise and developer environments.

Mistral AI's Mistral Small 4 packs 119 billion parameters and 128 experts, blending fast text responses, logical reasoning, and image processing—punching above its compact weight.

Trend alert: Open-source local inference tools like llama.cpp, Ollama, and vLLM are winning by ditching vendor lock-in.

• Ultimate customization:...

Qwen3.5 0.8B Reasoning unavailable on any benchmarked API providers – as an open weights model, self-hosting is the way to go for reasoning tasks.

Unsloth Studio streamlines local LLM workflows with no-code tools:

• Build synthetic data pipelines via NVIDIA NeMo Data Designer
• Generate data...

Breakthrough in compression: Applies structured pruning (neurons/heads/layers) and quantization (16-bit to 4/8-bit) to Mistral Large/Small, retaining...

Llama 4, Meta's fourth-generation open-weight large language models released in April 2025, now has a complete beginner-to-advanced tutorial. It's the most-downloaded open-source model family.

NVIDIA Nemotron 3 Releases

• 🔥 Nemotron 3 Expansion: Nvidia expands its open AI model portfolio with the Nemotron 3 lineup, including Nemotron 3...

Nemotron 3 lineup launches with multimodal models like Ultra (frontier performance, 5x throughput in NVFP4), Omni (audio/vision/language), and...

NVIDIA's latest open beast: Nemotron 3 Super delivers state-of-the-art reasoning, coding, and multimodal performance—fully downloadable and...

New YouTube tutorial breaks down LLM fine-tuning for specialized knowledge:

• Visuals + analogy: Library metaphor explains fine-tuning vs....

Microsoft launches Azure Fireworks AI for open models inside Azure AI Foundry, gaining momentum in open-model infrastructure. Key for enterprise deployment of open weights like Llama and Mistral.

Open Weights Rankings

• 🔥 GLM-5 Reasoning Leads: GLM-5 (Reasoning) is the highest-ranked open weights model with an Intelligence Index score of...

Open weights models are holding strong in key metrics:

• Qwen2.5 Instruct 72B GLM-5 (Reasoning) ranks highest among open weights with an Intelligence...