AI Briefing#open-weights
The latest AI and developer news about #open-weights, with the original source and a short summary.
The latest AI and developer news about #open-weights, with the original source and a short summary.
A refusal circuit with a sparse gate-amplifier structure was repeatedly confirmed across 12 open-weight models.
China's Z.ai has released GLM-5.1, an open-weight model with 754B parameters.
OpenAI has released a prompt-based safety policy package to help developers build safe AI services for teens.
Qwen3.5-397B-A17B has been released, a native multimodal model in which only 17B of the 397B parameters are activated.
Qwen3-Coder-Next targets SWE-Bench Pro-level performance with just 3B active parameters.
OpenAI has released gpt-oss-safeguard, an open-weight model that uses reasoning to interpret developers' custom safety policies.
OpenAI has released high-performance open-weight models, pushing forward the democratization of AI technology and the expansion of a US-led AI ecosystem.
OpenAI has released gpt-oss-120b and gpt-oss-20b, open-weight reasoning models under the Apache 2.0 license.
OpenAI has released gpt-oss, an open-weight model that supports powerful reasoning performance and efficient deployment.
Google has released Gemma 3, a new open model supporting multimodality and a 128k context.
Cohere For AI has released the 'Aya Vision' series, an open-weight multimodal model supporting 23 languages.
Cohere For AI unveiled the Aya Expanse 8B and 32B models, which maximize multilingual performance.
Qwen's next-generation audio model, **Qwen2-Audio**, has been released, supporting both voice and text input.
Stable Audio Open, an open-weight text-to-audio model trained on Creative Commons data, has been released.
Mistral AI has released Codestral, a 22B-scale open-weight code generation model supporting over 80 programming languages.
AI21 has released Jamba, a Mamba-based model supporting a 256K context window.
OpenAI argued that both open-weight releases and API-based delivery are necessary for the safe advancement of AI.
Mistral AI has released Mixtral 8x7B, an open-weight model that outperforms Llama 2 70B.