Product & Tech Updates

Anthropic launched Reflect, a beta dashboard (Free, Pro, and Max users with Memory on) that visualises Claude usage over 1-12 months - top topics, task types, peak hours - and coaches you via its 4D AI Fluency Framework (delegation, description, discernment, diligence), suggesting features like Projects or custom skills based on your patterns. It ships wellbeing controls (quiet hours, break nudges, reflection prompts) built with MIT Media Lab and Boston Children's Hospital experts, and excludes health-linked conversations entirely. TechCrunch's sharp read: beneath the mindfulness framing, Reflect showcases how much of your work runs through Claude - a retention play as much as a wellness one.

AI startup Decart launched Oasis 3, a real-time interactive "world model" that generates photorealistic driving environments, available via API at $0.02 per second. It's initially aimed at autonomous-vehicle testing of rare scenarios, with robotics and other physical-AI uses next. The model impresses on photorealism and runs for hours, but degrades over time and doesn't yet simulate physics - cars drive straight through each other.

Apple's long-delayed Siri AI finally arrived at WWDC 2026 - a rebuilt assistant with multi-turn conversation, cross-app actions, and its own app - but two footnotes matter most. It's powered partly by Google's Gemini models, an admission Apple couldn't close the frontier-model gap alone; and the initial beta is English-only, excludes China outright, and won't reach EU iPhones at launch. It lands as Tim Cook's final WWDC before John Ternus takes over.

Anthropic launched Claude Fable 5 - a Mythos-class model it bills as state-of-the-art across software engineering, knowledge work, vision, and science - alongside Claude Mythos 5, the same model with cyber safeguards lifted for vetted defenders in Project Glasswing. To release that power safely, Fable 5 routes sensitive cyber, biology/chemistry, and distillation requests to Opus 4.8, with fallback triggering in under 5% of sessions, and adds a 30-day data-retention rule. Both ship at $10/$50 per million input/output tokens - less than half the price of Mythos Preview - though a US government directive suspended access just three days later.

At WWDC 2026 - Tim Cook's last as CEO before handing off to John Ternus - Apple leaned into catching up on AI, unveiling a more capable, conversational Siri (now a standalone app) powered in part by Google's Gemini. The next wave of Apple Intelligence adds cross-app context, AI reply suggestions, natural-language Shortcuts, and new Photos editing tools. iOS 27 stretches back to the iPhone 11 with sizable performance gains, plus App Store subscription bundles and personalized recommendations.

Google cut its Google AI Plus subscription from $7.99 to $4.99 a month while doubling included storage to 400GB - dragging a price war that had been simmering in emerging markets into the US. Analysts frame it as the start of a commoditization era, where Google's vertical integration and distribution erode margins for pure-play AI providers. Notably, Anthropic hasn't matched the move with any budget tier or localized pricing.

OpenAI is rolling out its most capable ChatGPT memory system yet, built on a background process called dreaming that synthesizes memories for freshness, accuracy, and scale. It targets the staleness and correctness problems that show up across hundreds of millions of users and multi-year histories. It's live for Plus and Pro in the US now, with Free and Go users following over the coming weeks after a roughly 5x cut in serving cost.

Anthropic added two pieces to its Claude Partner Network: a tiered Services Track (Select, Preferred, Global Premier) that ranks firms by certified staff, production deployments, and public customer stories, and a Partner Hub portal where partners track their standing daily and customers find qualified firms. Since March, 40,000+ firms have applied and 10,000+ consultants have earned Claude certifications. The network is backed by a $100M investment in training, support, and marketing.

OpenAI updated GPT-Rosalind, its life-sciences model, pairing GPT-5.5's agentic coding and tool use with stronger intelligence in drug-discovery domains like medicinal chemistry and genomics. New expert-judged benchmarks show gains over GPT-5.5 with notably fewer tokens, and two Codex plugins plus interactive viewers turn reasoning into executable bioinformatics workflows. Access expands globally through a trusted-access structure, with Novo Nordisk among early named partners.

OpenAI launched Rosalind Biodefense, a program that sponsors trusted developers to build biodefense and pandemic-preparedness tools on GPT-Rosalind, and expanded trusted access to the model for select US government and allied public-health partners. The framing is "defensive acceleration" - making sure frontier AI advantages those preventing, detecting, and responding to biological threats, paired with layered safeguards. Launch partners span DNA-synthesis screening, national labs, Johns Hopkins APL, and CEPI.

Anthropic released Claude Opus 4.8, an upgrade to its Opus class with stronger coding, agentic, and knowledge-work performance at the same price ($5/$25 per million input/output tokens). New features include user-controllable effort levels, a Claude Code "dynamic workflows" mode that runs hundreds of parallel subagents, and fast mode now 3x cheaper. Anthropic highlights improved honesty - the model is around 4x less likely than its predecessor to let flaws in its own code pass unremarked.

Google is extending Search into an agentic experience, introducing information agents - personalized AI agents that work in the background 24/7 to find what you need and help you act - rolling out over the summer to AI Pro and Ultra subscribers. It's also infusing Search with agentic coding, so Gemini 3.5 Flash and Antigravity build custom, generative interfaces (dynamic layouts, interactive visuals, even persistent dashboards) for individual queries, free for everyone. The move builds on AI Mode, which passed 1 billion monthly users in a year.

Google announced Gemini Omni, a new model family capable of generating output in any modality from any input - combining Gemini's intelligence with Google's generative media models in what it frames as a leap in world understanding. The first release, Gemini Omni Flash, starts with video outputs (with image and text to follow) and is available today in the Gemini app, Google Flow, and YouTube Shorts. API access for developers and enterprises follows in the coming weeks.

Google unveiled Gemini Spark, a personal AI agent inside the Gemini app that takes action on a user's behalf and runs 24/7 on dedicated Google Cloud virtual machines, so it works in the background without a laptop open. Powered by Gemini 3.5 and Google's Antigravity harness, it handles long-horizon tasks and connects to third-party tools via MCP. It began rolling out to trusted testers, with a beta reaching Google AI Ultra subscribers in the US shortly after.

Chinese tech giants are shipping and teasing new model generations at a pace that's starting to feel like a platform land-grab, not just R&D cadence. Video generation is a particular battleground, with ByteDance-linked efforts drawing attention for realism claims.

OpenAI introduced a production model running on Cerebras hardware, optimized for near-instant coding interactions and reported to exceed 1,000 tokens/sec in output speed. It's a quiet but meaningful platform shift: inference performance and cost are pushing major AI vendors to diversify beyond Nvidia.