AI News
AI community posts, news and release updates
Latest updates
A user laments that ChatGPT's writing has now surpassed Claude, expressing disappointment.
- Claude
- ChatGPT
- writing
- Launch
Qwen-Audio-3.0-TTS: 16 Languages, Fine-Grained Control over Whisper/Anger/Laughs, #1 on TTS Leaderboard
Published Jul 23, 2026Alibaba released Qwen-Audio-3.0-TTS, the latest text-to-speech model with Flash (real-time) and Plus (high-quality) variants. It features fine-grained inline tags for whisper, anger, breaths, and laughs, natural-language control (e.g., 'read this slowly like a bedtime story'), 16 languages, and up to 3-minute one-pass generation. It is now #1 on the Artificial Analysis TTS Leaderboard.
- Qwen
- TTS
- audio
- leaderboard
- Speech Synthesis
A counterexample to the Dinitz–Garg–Goemans conjecture was discovered using ChatGPT, showcasing AI's capabilities in theoretical computer science.
- ChatGPT
- AI Research
- mathematics
- conjecture
- Launch
Qwen 3.8 Max Preview Tested: Builds Three.js Scenes in One Shot, Faster with Same Quality
Published Jul 23, 2026AI/ML API pitted Qwen 3.8 Max Preview against other top models, each building Three.js scenes (war galley, Trojan horse, Greek helmet) in one shot. Qwen completed the task considerably faster with no visible quality loss. The model is coming soon to AI/ML API.
- Qwen
- Three.js
- AI/ML API
- model benchmark
- Launch
Astribot Launches Lumo-2: A World-Action Model on Frozen Qwen VLM, Robot Learns Latte Art with Thin Adapters
Published Jul 23, 2026Chinese robotics company Astribot released Lumo-2, a World-Action Model based on a frozen Qwen-3.5 4B VLM, trained in three progressive stages: aligning actions with latent world dynamics, aligning with vision-language, and end-to-end training. While not top on benchmarks, its novelty lies in combining latent world modeling with action generation and leveraging unannotated human video for generalization. With only thin trainable adapters, it's suitable for real-time embedded inference and excels at long-horizon tasks.
- Qwen
- world model
- Embodied AI
- robot
- action model
- Astribot
- Lumo-2
- Idea
Liang Wenfeng's 4-Hour Interview: DeepSeek's AGI Path, China-US AI Gap, and Cost Advantage
Published Jul 23, 2026DeepSeek founder Liang Wenfeng detailed his AGI vision in a 4-hour interview: the ultimate goal is AGI, not short-term profit; the tech roadmap is language model → CoT → Agent → continual learning → self-iteration → embodied AI. He estimates China trails the US by 6-18 months overall but holds advantages in efficiency and cost, potentially forming a 'systemic low-cost' advantage akin to Chinese manufacturing. He also shared views on NVIDIA, Huawei chips, OpenAI, and talent competition.
- 开源
- AGI
- DeepSeek
- 成本优势
- 中美AI竞争
- 梁文锋
- 技术路线
Signs indicate Anthropic is preparing to release Claude Opus 5. Timeline not yet confirmed, but such preparations usually precede a launch. The Opus release cadence has been 42-70 days; it's been 56 days since last version.
- Anthropic
- release
- Claude Opus 5
NInfer demonstrates impressive performance for Qwen3.6-35B-A3B on a single RTX 5090: 271 tok/s, doubled to 543 tok/s with speculative decoding.
- Qwen
- RTX 5090
- 推理性能
Rumor says Claude Opus 5 launches today with Fable 5-level intelligence at half the price. Opus 4.8 had fallen behind GPT 5.6 and open-source Kimi K3. Anthropic is about to answer.
- Anthropic
- Fable 5
- Claude Opus 5
A developer shares experience building a cyberpunk FPS using Grok Build, Grok 4.5 with Unity and Blender, working until 5 AM and waiting for the new Grok App Deployer.
- Grok
- 游戏开发
- AI辅助
- Unity