HuggingFace Releases Jbliterated Models with Refusal Removed
HuggingFace announced on July 20, 2026, the release of Jbliterated models that remove refusal behavior without affecting personality or knowledge.
Model Release
459 articles
New AI models as they ship — large language models, multimodal systems, and open-weight releases. Capabilities, context windows, pricing, and availability, summarized so you can compare what is actually new.
LATEST NEWS
-
-
OpenAI Launches Health Feature in ChatGPT
OpenAI is rolling out a health feature in ChatGPT for U.S. users aged 18 and older, offering better advice to paying subscribers.
-
Black Forest Labs Launches Flux 3 With Native Audio Video Generation
Black Forest Labs has released Flux 3, a multimodal model that generates videos with native audio up to 20 seconds long, marking a new milestone in AI capabilities.
-
OpenAI Expands ChatGPT Health Access to All U.S. Users
OpenAI announced ChatGPT Health is now available to all U.S. users over 18, following a lawsuit over a dangerous health suggestion. The feature now integrates with medical records and health platforms.
-
Anthropic Enhances Claude Voice Mode With New Models
Anthropic announced an update to Claude's voice mode, offering Opus, Sonnet, and Haiku models for more capable interactions.
-
Health in ChatGPT Launches for U.S. Users
OpenAI rolls out Health in ChatGPT to U.S. users, allowing secure connections to Apple Health and medical records. Over 300 million people use ChatGPT for health-related questions weekly.
-
Google's Gemini Reaches 950M Monthly Users
Google's Gemini AI assistant now has over 950 million monthly users, tripling its user base from last year, according to the company.
-
Grok 4.5 Now Available on iOS, Android, Web, and X
X.ai released Grok 4.5, its most capable model yet, on July 22, 2026, across iOS, Android, web, and X platforms.
-
Banaxi-Tech Targets Arithmark 2 Weakness on Open SLM Leaderboard
Banaxi-Tech and AxiomicLabs are addressing a flaw in the Open SLM Leaderboard's ranking system after specialist models exploited it to gain high scores.
-
Google CEO Pichai: Gemini's Next Leap Relies on Larger Base Models
Google CEO Sundar Pichai said Gemini's next leap depends on building 'much larger base models,' as the app reached 950 million monthly active users in Q2 2026.
-
Poolside Unveils Laguna S 2.1, Small Open-Weight Coding Model
Poolside's Laguna S 2.1, a small open-weight coding model, outperforms larger models on Terminal-Bench 2.1 with a 70.2% score, according to the company.
-
White House Accuses Moonshot of Copying Anthropic's Fable in Kimi K3
White House science advisor Michael Kratsios claims Moonshot, the Chinese firm behind Kimi K3, copied Anthropic’s Fable LLM, sparking debate over tech theft in AI development.