AI Models
29 articles
Alibaba Launches Qwen3.8-Max With 2.4 Trillion Parameters, Claiming It Trails Only Anthropic's Fable 5
Alibaba's Qwen team unveiled a 2.4-trillion-parameter Qwen3.8-Max preview, its first trillion-parameter multimodal model, saying it trails only Anthropic's Fable 5 as it moves to blunt Kimi K3's momentum.
German Consortium Releases Soofi S, an Open 30B Model That Tops English and German Benchmarks
A German consortium released Soofi S 30B-A3B, a hybrid MoE model trained on Deutsche Telekom's Munich cloud that tops fully open models on English and German benchmarks.
Moonshot AI's Kimi K3: 2.8-Trillion-Parameter Open Model Challenges Claude and GPT
China's Moonshot AI released Kimi K3, a 2.8-trillion-parameter open model that trails only Claude Fable 5 and GPT-5.6 Sol and tops the Frontend Code Arena.
NVIDIA Releases Nemotron 3 Embed, an Open Embedding Collection With an 8B Model Topping the RTEB Leaderboard
NVIDIA's open Nemotron 3 Embed collection targets production-scale RAG and agentic retrieval, with its 8B checkpoint ranking #1 on the RTEB benchmark.
Moonshot AI Unveils Kimi K3, a 2.8-Trillion-Parameter Open-Weight Model Challenging Top US Rivals
China's Moonshot AI announced Kimi K3, the world's largest open-weight model at 2.8 trillion parameters, nearly matching GPT-5.6 and Claude Fable 5 at a fraction of the cost.
Alphabet Shares Fall 4% as Google Delays Gemini 3.5 Pro Again Over Coding Shortfalls
Google's flagship Gemini 3.5 Pro model is months behind schedule as coding performance falls short of internal goals, sending Alphabet shares down 4% as rivals pull ahead.
Mira Murati's Thinking Machines Launches Inkling, a 975-Billion-Parameter Open-Weight AI Model
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, released its first model: a 975-billion-parameter open-weight model designed for customization over raw benchmark dominance.
OpenAI's GPT-5.6 Sol Deletes User Files and Databases, Sparking Safety Concerns
Developers report GPT-5.6 Sol deleting files, databases, and entire directories without permission. OpenAI's own system card warned of the risk before launch.
Meta Launches Muse Spark 1.1 With Aggressive API Pricing, Igniting an AI Coding Price War
Meta's first paid AI model, Muse Spark 1.1, debuts at $1.25 per million input tokens with a 1M context window, undercutting OpenAI, Anthropic, and xAI in the coding market.
OpenAI Launches GPT-Live: Full-Duplex Voice Models That Listen and Speak at Once
OpenAI's new GPT-Live voice models let ChatGPT listen and speak simultaneously using a full-duplex architecture. GPT-Live-1 is live for paid users now.
Refiant Launches Protea, a 10 Million-Token Context Window AI Model, No Waitlist Required
South African-founded Refiant has launched Protea, a suite of long-context AI models with a 10 million-token context window available immediately with no waitlist, targeting enterprise codebases and regulatory archives.
Mistral AI Launches Robostral Navigate, Its First Robotics Model for Single-Camera Robot Navigation
Mistral AI's Robostral Navigate is an 8B model that enables robots to autonomously navigate complex environments using only a single RGB camera, achieving 76.6% on the R2R-CE benchmark.
OpenAI Gets Green Light to Release GPT-5.6 to Everyone on July 9
The US Commerce Department cleared OpenAI to publicly release GPT-5.6 Sol, Terra, and Luna on July 9, ending weeks of government-requested limits.
SpaceXAI Launches Grok 4.5 in Joint Effort With Cursor, Challenging Anthropic and OpenAI
Elon Musk's SpaceXAI released Grok 4.5, a 1.5-trillion-parameter model trained with Cursor data, claiming Opus-class performance at lower cost. The launch marks xAI's first major release under its new SpaceXAI brand.
Meta Launches Muse Image, Its First In-House AI Image Generator, Inside Instagram and WhatsApp
Meta launched Muse Image and Muse Video on July 7, 2026, its first in-house AI image generator, built into Instagram, WhatsApp and the Meta AI chatbot.
Portugal Launches Amália, Its First Open-Source AI Model, in a Push for European AI Sovereignty
Portugal has released Amália, its first open-source national AI model, built for European Portuguese as public digital infrastructure. Here is what it does and who built it.
Meta's 'Watermelon' AI Model Matches GPT-5.5 on Benchmarks as Zuckerberg Flags Agent Delays
Meta AI chief Alexandr Wang says the next model, Watermelon, has caught up with OpenAI's GPT-5.5 using 10x more compute, even as Zuckerberg admits AI agents are behind schedule.
Google Ships Nano Banana 2 Lite and Gemini Omni Flash for Faster, Cheaper Media Generation
Google released Nano Banana 2 Lite, its fastest image model at $0.034 per 1,000 images, alongside Gemini Omni Flash for video generation and conversational editing on June 30, 2026.
Anthropic Launches Claude Sonnet 5, Its Most Agentic Mid-Tier Model Yet
Anthropic released Claude Sonnet 5 on June 30, 2026, a mid-tier model with agentic performance rivaling Opus 4.8 at far lower cost, with intro pricing of $2 per million input tokens.
Meituan Open-Sources LongCat-2.0, a 1.6 Trillion-Parameter Coding Model Trained Entirely on Chinese Chips
China's Meituan released LongCat-2.0, a 1.6 trillion-parameter Mixture-of-Experts coding model trained on 50,000 domestic chips without Nvidia hardware.
Claude Opus 4.8 Fast Mode Arrives in GitHub Copilot as Anthropic Pushes Deeper Into Coding
GitHub Copilot now offers Claude Opus 4.8 fast mode in preview, giving developers a 2.5x speed boost at a fraction of previous fast-mode costs.
Europe Picks Milan's Domyn to Build a 400-Billion-Parameter Open AI Model
The EU has selected the EUROPA consortium led by Italian startup Domyn to build a roughly 400-billion-parameter, open-weight frontier AI model covering all 24 official EU languages.
OpenAI Launches GPT-5.6 Sol, Terra, and Luna in Limited Preview With New Safety Guardrails
OpenAI unveiled the GPT-5.6 family led by flagship Sol, with Terra and Luna tiers, new multi-agent reasoning, and its most robust safety stack yet in a staggered preview.
Sakana AI Launches Fugu, a Multi-Agent System That Rivals Frontier Models by Orchestrating Rivals
Tokyo startup Sakana AI unveiled Fugu, a single-API orchestrator that dynamically routes tasks across a swappable pool of frontier LLMs to match Anthropic's Fable 5.
Google Adds Built-In Computer Use to Gemini 3.5 Flash for Cross-Platform AI Agents
Google DeepMind made computer use a built-in tool in Gemini 3.5 Flash, with adversarial training and enterprise safeguards to curb prompt-injection risk.
Microsoft Unveils Seven In-House MAI Models With Zero Distillation to Cut Costs and Reduce OpenAI Reliance
Microsoft launched seven MAI models at Build 2026, including a reasoning model that matches Anthropic's Opus 4.6 on coding, all trained from scratch with zero distillation.
OpenAI Prepares GPT-5.6 With 1.5M Token Context Window and Pro Variant
OpenAI is set to launch GPT-5.6 as soon as next week, with a 1.5 million token context window, improved long-horizon coding, and Mini and Pro variants targeting Anthropic.
Chinese AI Models Overtake U.S. Rivals in Global Token Usage as DeepSeek V4 Takes the Top Spot
OpenRouter data shows Chinese models like DeepSeek V4 Flash and MiniMax M3 now dominate global token consumption, leaving ChatGPT and Gemini outside the top 10.
Zhipu Launches GLM 5.2 as CEO Spars With Elon Musk Over China's Frontier AI Timeline
Chinese AI lab Zhipu released GLM 5.2 as CEO Tang Jie pushed back on Elon Musk's Q1 2027 timeline for China to reach Fable-class AI.