GPT-6 Astra tops Perplexity's research benchmark, beating Claude Fable 5.1 by 13.5%
Perplexity's own research-agent benchmark now crowns GPT-6 Astra, which scores 13.5% higher than Claude Fable 5.1 — while costing 6.1% less per task.
Perplexity
Most Recent 
Israel's state-funded 'think tank' is gaming AI search citations
Tencent's cloud chief says a 'seriously insufficient' compute supply has held back its Hunyuan AI models and products, but argues endurance matters more than an early lead in the AI race.
Nvidia Mulled Billions for Perplexity's Tech. Now It's Investing Instead.
Yahoo is rolling out AI answer engine Scout this year, using Anthropic’s Claude and Bing to win back Gen Z users.
More Search Rounds Beat Switching Engines, OpenRouter's AI Benchmark Shows
DeepSeek open-sourced Harness and revealed the Cordis design paper, explaining how making every component — including the agent loop — a swappable plugin paves the way toward recursive self-improvement.
Meta's AI search crawlers are so aggressive they're tripping indie devs' server alarms
An OpenAI product lead showed developers how to swap GPT into Claude Code's harness. One user got banned for it, and the aftermath turned into a very public — and very petty — cross-company spat.
Amazon Kills Most Nova AI Models, Hands Reins to OpenAI Mentor
A Claude Max 20x pricing bug allowed free access to a $200/month plan, leading to mass account bans that also swept up paying customers and long-time users.
Commercial AI models refused to analyze a real hack
Seed Audio 1.0 generates up to 2 minutes of audio with 100ms time control, supports 20+ languages, and maintains consistent character voices. Available now via API.
Even the Best Research Agent Only Gets 1 in 7 Objects Right
Mira Murati's Thinking Machines Lab releases Inkling, a 975B-param MoE model that uses a fraction of tokens compared to Nvidia's Nemotron.
Sakana AI’s Self-Aware Bricks Detect Their Own Damage Without a Central Brain
The National Science Foundation is set to ban its grantees from collaborating with Chinese researchers on most projects, potentially freezing a major channel of US-China scientific cooperation.
Perplexity is building a Cursor rival called Teammate
Cognition's new SWE-1.7 coding AI model, based on Kimi K2.7 Code and running on Cerebras chips, outputs 1,000 tokens per second and can handle tasks lasting up to 6 hours.