Source URL: https://simonwillison.net/2025/Jul/31/more-models/
Source: Simon Willison’s Weblog
Title: More model releases on 31st July
Feedly Summary: Here are a few more model releases from today, to round out a very busy July:
Cohere released Command A Vision, their first multi-modal (image input) LLM. Like their others it’s open weights under Creative Commons Attribution Non-Commercial, so you need to license it (or use their paid API) if you want to use it commercially.
San Francisco AI startup Deep Cogito released four open weights hybrid reasoning models, cogito-v2-preview-deepseek-671B-MoE, cogito-v2-preview-llama-405B, cogito-v2-preview-llama-109B-MoE and cogito-v2-preview-llama-70B. These follow their v1 preview models in April at smaller 3B, 8B, 14B, 32B and 70B sizes. It looks like their unique contribution here is “distilling inference-time reasoning back into the model’s parameters" – demonstrating a form of self-improvement. I haven’t tried any of their models myself yet.
Mistral released Codestral 25.08, an update to their Codestral model which is specialized for fill-in‑the‑middle autocomplete as seen in text editors like VS Code, Zed and Cursor.
And an anonymous stealth preview model called Horizon Alpha running on OpenRouter was released yesterday and is attracting a lot of attention.
Tags: llm-release, openrouter, mistral, generative-ai, cohere, ai, llms
AI Summary and Description: Yes
Summary: The text discusses several recent developments in the realm of large language models (LLMs) and generative AI, highlighting new models and their capabilities, particularly in the context of accessibility and innovative features presented by different companies. This information is pertinent for professionals in AI, as it reveals trends in model development and deployment.
Detailed Description:
The text serves as an update on various recent LLM releases, showcasing advancements in generative AI. Each release emphasizes unique features and capabilities, which may have implications for security, performance, and compliance within AI-driven applications. Here’s a detailed breakdown of the main points:
– **Cohere’s Command A Vision**:
– This is their first multi-modal LLM capable of processing image inputs.
– It operates under open weights published under Creative Commons Attribution Non-Commercial, requiring licensing for commercial use.
– The availability of open weights can facilitate research and development but raises potential compliance considerations for commercial applications.
– **Deep Cogito’s Hybrid Reasoning Models**:
– Deep Cogito has released four hybrid reasoning models, enhancing their previous offerings.
– Notable versions include sizes ranging from 70B down to 3B parameters.
– Unique feature: “distilling inference-time reasoning back into the model’s parameters,” which indicates an innovative approach to model adaptation and improvement.
– This self-improvement methodology presents implications not just for AI performance but for oversight and governance as well.
– **Mistral’s Codestral Update**:
– An enhancement to the Codestral model, specifically designed to improve autocomplete functionality in text editing environments.
– The focus on utility in widespread applications like code editing suggests that security updates and compliance may be necessary to protect sensitive data being handled within such tools.
– **Horizon Alpha**:
– Introduced as an anonymous preview, this model is drawing significant attention, suggesting potential high interest within the AI community.
– The lack of disclosure around its nature may provoke questions regarding security, ethical usage, and compliance, particularly if “stealth” implies proprietary or sensitive features.
Overall, these developments reflect the ongoing innovation in the AI field, particularly in the creation and refinement of LLMs. Professionals should consider how these advancements intersect with security and compliance frameworks, potentially influencing best practices in AI deployment and use.