Alibaba Releases Qwen3.8, Its First Qwen-Max-Class Open-Weight Model
Alibaba's Qwen team released Qwen3.8 in August 2026, its first Qwen-Max-class open-weight model: a 2.4T MoE with 95B active parameters, followed by a 27B model.
Open-Source AI
News about open-weight and open-source AI models and tools, including Llama, DeepSeek, Qwen, Mistral, Gemma and gpt-oss, and their impact on developers.
10 stories
Open-weight models have narrowed the gap with proprietary frontier systems. You can now run capable models on your own hardware, fine-tune them and deploy them without per-token fees. This category covers the most important open releases and licensing changes, and what they mean for self-hosted and local AI in .NET.
Moonshot AI Releases Kimi K3 Weights, a 2.8-Trillion-Parameter Open Model
Moonshot AI released Kimi K3's weights in July 2026, a 2.8T-parameter multimodal MoE model with a 1M-token context and new terms for large commercial users.
DeepSeek V4 Released: Open-Weight 1.6T MoE With a 1M-Token Context Built for Agents
DeepSeek released V4-Pro and V4-Flash in April 2026, MIT-licensed open-weight MoE models with a 1M-token context and far cheaper long-context inference.
Google Releases Gemma 4 Open Models Under Apache 2.0, From Phones to a Single GPU
Google DeepMind released Gemma 4 in April 2026, multimodal open models from E2B to 31B, and moved Gemma to the Apache 2.0 license for the first time.
Mistral Small 4: Apache 2.0 Model Unifies Reasoning, Coding and Vision
Mistral AI released Mistral Small 4 in March 2026, an Apache 2.0 open-weight 119B MoE model with 6.5B active parameters that unifies chat, reasoning and coding.
OpenAI Releases gpt-oss-120b and gpt-oss-20b, Its First Open-Weight Models Since GPT-2
OpenAI released gpt-oss-120b and gpt-oss-20b in August 2025, its first open-weight language models since GPT-2, under Apache 2.0 for reasoning and tool use.
Moonshot AI Open-Sources Kimi K2, a 1-Trillion-Parameter Agentic Model
Moonshot AI open-sourced Kimi K2 in July 2025, a 1-trillion-parameter MoE model with 32B active parameters built for tool use, under a modified MIT license.
Alibaba Releases Qwen3: Apache 2.0 Open Models with Hybrid Thinking Modes
Alibaba's Qwen team released Qwen3 in April 2025: eight Apache 2.0 open-weight models, from 0.6B to a 235B MoE, that switch between thinking and fast modes.
Meta Releases Llama 4 Scout and Maverick, Its First Open-Weight MoE Models
Meta released Llama 4 Scout and Maverick in April 2025, its first natively multimodal mixture-of-experts open-weight models, with a 10M-token context claim.
DeepSeek-R1 Release: MIT-Licensed Open Reasoning Model Rivals OpenAI o1
DeepSeek-R1 arrived in January 2025 as an MIT-licensed open-weight reasoning model that DeepSeek says matches OpenAI o1, and it shook AI markets within a week.
No matches. Try the site search.