AI moves fast. Here's what we're reading this week.
Multiple sources at OpenAI hint at a summer 2026 release for GPT-5, with first-class video input and an output mode targeted at consumer creators. The bigger story may be the pricing — early indications suggest GPT-5-mini will undercut current Haiku-tier rates.
Claude Sonnet 4.5 is shipping with a 1M-token context window for select customers. Early benchmarks show meaningful gains on long-document reasoning, particularly for codebases over 200K tokens. Pricing remains at $3 input / $15 output per million.
After two years of AI-assisted content, the brands that win aren't producing more — they're producing fewer but more specific assets. A breakdown of how 6 mid-market teams cut their volume by 40% and grew engagement.
Gemini 2.5 Pro now tops the SWE-Bench Verified leaderboard, edging out Claude and GPT-4. The model also shows surprising gains on agentic tasks. Whether it changes the developer landscape depends largely on Google's go-to-market.
A 12-line prompt template — turning user inputs into a cinematic scene description for image generators — has racked up 40M views across TikTok this month. Its creator, an art student in Berlin, says she didn't expect it to leave her group chat.
Excel users can now type natural-language descriptions and have them converted to formulas inline. Power users complain about latency; spreadsheet beginners report dramatic productivity gains. Adoption is strongest in finance and HR teams.
DeepSeek-V3 and Qwen-3-Math are now within 4 percentage points of GPT-4 on MATH-500. The gap on coding tasks remains wider, but the trajectory has analysts revising 2026 forecasts.
After 18 months of policy churn, most US school districts surveyed have shifted from outright bans to redesigning assignments for an AI-assumed world. The shift mirrors how calculators were absorbed in the 1990s.
OpenAI's browser-using agent product hit 53% on a curated set of real-world tasks in independent testing. The remaining failures cluster around payment forms and CAPTCHAs — both, notably, by design.
Power users are increasingly using multi-turn workflows where the model reflects, refines, and asks for clarification. The shift is leaving single-prompt productivity hacks behind. A look at what's replacing them.
Notion's latest update lets users embed prompt templates inside individual databases — turning each row into an AI-powered button. Teams using it for hiring pipelines and customer feedback report sharp time savings.
Recruiters increasingly use detectors that flag AI-typical phrasings ("meticulously crafted", "passionate about leveraging"). Candidates who use AI for editing rather than drafting still get through. Candidates who copy-paste do not.
