AI 新闻 —— July 24, 2026
PulseAugur 当天浮现的 20 条头条故事 —— 综合实验室、论文及开发者社区的信号进行排序。
-
OpenAI launches GPT-5.6 with tiered Sol, Terra, and Luna models
OpenAI has launched GPT-5.6, a new model family that deviates from its previous single-model approach by offering three distinct tiers: Sol, Terra, and Luna. Sol is designed for complex reasoning and coding tasks, positioned as a cybersecurity model and powering ChatGPT Work. Te…
-
Anthropic releases Claude Opus 5, focusing on economics
Anthropic has launched its latest model, Claude Opus 5, on July 24, 2026. This marks the company's fourth model release in less than two months, following the introduction of Mythos 5. The article suggests that the primary focus of this new release is on its economic implication…
-
Anthropic releases Opus 5, matching Fable 5 performance at lower cost
Anthropic has released Opus 5, a new version of its flagship large language model, which offers performance close to its top-tier Fable 5 model but at half the price. The company states that Opus 5 excels in knowledge work, complex coding, and scientific research, showing improv…
-
Chinese youth AI team releases model rivaling Claude Opus 4.8
A Chinese team of developers has released a new AI model that reportedly rivals the performance of Anthropic's Claude Opus 4.8. The model's capabilities are being closely watched as it represents a significant advancement from a young, emerging team in the competitive AI landscape.
-
Pika Labs launches Pika 2.5 with enhanced physics and sound effects
Pika Labs, a company founded by two Stanford graduates, has released its flagship video generation engine, Pika 2.5. This new version, available by default on all plans in 2026, boasts improved physics simulation, allowing for realistic object interactions like breaking glass. W…
-
Black Forest Labs launches FLUX 3 Video, outperforming Gemini Omni and Grok Imagine
Black Forest Labs has released FLUX 3 Video, a multimodal flow model that reportedly surpasses existing models like Gemini Omni and Grok Imagine in video generation capabilities. The model offers a wide range of features including text-to-video, image-to-video, video-to-video ge…
-
Claude Fable 5 leads AI race, challenged by GPT-5.6 Sol and new Beijing model
As of July 2026, the AI model race is exceptionally competitive, with several leading contenders vying for the top spot. Claude Fable 5 currently holds a slight edge in independent evaluations due to its advanced reasoning, coding, and long-context abilities. GPT-5.6 Sol is a cl…
-
Thinking Machines unveils Inkling, an open-weight audio-native LLM
Thinking Machines, a lab founded by former OpenAI CTO Mira Murati, has released Inkling, an open-weight model with native audio processing capabilities. This 975 billion parameter model, with 41 billion active parameters, can process text, images, and audio directly without exte…
-
OpenAI upgrades free ChatGPT to GPT-5.6 Terra model
OpenAI has upgraded the free tier of ChatGPT to utilize the GPT-5.6 Terra model, offering performance comparable to the previous flagship GPT-5.5. This change, effective July 9, 2026, provides users with enhanced capabilities at no additional cost, though direct access from Russ…
-
Alibaba Cloud releases Qwen2 model series with long-context and multilingual capabilities
Alibaba Cloud has released the Qwen2 series of open-source language models, offering a range of sizes from 0.5 billion to 72 billion parameters, including a mixture-of-experts model. These models boast enhanced capabilities in long-context reasoning, supporting up to 128K tokens…
-
Sber's GigaChat offers free, unlimited AI image generation with Kandinsky 6.0
Sber has integrated its Kandinsky 6.0 image generation model into its GigaChat AI assistant, offering free and unlimited image creation. This move, announced on April 28, 2026, makes advanced AI art generation accessible without requiring VPNs, foreign payment methods, or subscr…
-
Chinese Gen Z team develops AI model rivaling Opus 4.8, Meta's Facebook introduces user verification badges
A Chinese team has developed a new AI model that rivals the performance of Opus 4.8, a significant achievement for a team composed of individuals born after 2000. This development was noted by Liang Wenfeng, who is reportedly betting on such advancements. Separately, Meta's Face…
-
AssemblyAI's Universal-3.5 Pro Realtime hits human parity in speed and accuracy
AssemblyAI's Universal-3.5 Pro Realtime model has achieved a significant milestone by being the sole entry within Coval's Human Parity Zone on an independent speech-to-text leaderboard. This zone signifies models that match or surpass human performance in both accuracy and speed…
-
AMI Labs raises $1B for 'world models' amid LLM critique · 1 source tracked
Alexandre Lebrun, formerly of Meta and Nabla, has raised $1.03 billion for his new venture, AMI Labs, with Yann LeCun as chairman. Lebrun predicts "world models" will become a prevalent buzzword, but argues there's a fundamental architectural disagreement with current LLMs. He c…
-
Chinese youth team's AI model rivals Anthropic Opus 4.8 performance
A team of young Chinese developers has reportedly achieved performance levels comparable to Anthropic's Opus 4.8 model. This development, highlighted by investor Liang Wenfeng, suggests rapid progress in AI capabilities from emerging global players. The specific model or team na…
-
New research explores discrete flow matching and RL for generative models
Two research papers explore advancements in generative modeling, focusing on discrete structures and flow-based models. The first paper introduces context-weighted discrete flow matching to improve generation quality on discrete data by considering local context, achieving up to…
-
New AI Tool 'In the Weights' Measures Personal Recognition in LLM Parameters
A new service called "In the Weights" launched in June 2026, allowing users to check how well 13 different large language models recognize individuals based on their internal parameters rather than web searches. The tool, created by former OpenAI employees Thomas Dimson and Joey…
-
Anthropic launches Claude Opus 5 with double performance at same price
Anthropic has released Claude Opus 5, featuring a significant price reduction and performance improvements. The new model is available at the same price as its predecessor, Opus 4.8, but offers double the performance on certain benchmarks. It also demonstrates enhanced capabilit…
-
Few-shot prompting's effectiveness questioned in new LLM research
Recent research from 2025-2026 indicates that few-shot prompting, which involves providing examples in prompts, may no longer be as effective for advanced reasoning tasks with modern LLMs. Studies on models like Qwen2.5 and Qwen3 suggest that while few-shot examples can help ali…
-
Blue Funnel Line unveils FLUX 3 multimodal model, enabling robot control
Blue Funnel Line has introduced FLUX 3, a new multimodal foundation model that integrates image, video, and audio generation. This model has been further developed in collaboration with mimic robotics to create FLUX-mimic, a video-action model capable of controlling robots. FLUX…