AI News Second Week June 2026: ChatGPT Dreaming V3, Microsoft MAI Models, US Federal AI Bill and More
Category: AI News, Weekly Roundup, AI Tools
Quick Answer: The second week of June 2026 delivered six stories that will reshape how you use and build with AI. ChatGPT's Dreaming V3 launched on June 4, permanently changing how the tool remembers you. Microsoft unveiled seven proprietary MAI models at Build 2026 on June 2, including MAI-Code-1 now live in GitHub Copilot and VS Code. The US federal AI bill dropped 269 pages of legislation that would override every state AI law in America. xAI won a sweeping US government contract for Grok. OpenAI updated GPT-Rosalind for life sciences with access to 50-plus scientific databases. And for the first time, real market share data confirms that the single-platform era in AI is over: ChatGPT fell from 76.5% to 54.7% in 15 months while Claude grew 306% in a single quarter.
Last week was the biggest AI news week of the year so far. This week matched it.
Between June 2 and June 9, 2026, ChatGPT got its most significant architectural upgrade since launch, Microsoft declared independence from OpenAI with seven proprietary models, the US Congress dropped a 269-page AI bill that could freeze every state AI law in the country, and real market share data painted the clearest picture yet of who is actually winning the AI race.
None of these are minor product updates. Every one of them changes the landscape in a way that will still matter in a year. Here is the complete breakdown, with honest context on what each story actually means.
1. ChatGPT's Memory Just Got a Complete Overhaul
This is the story with the most immediate impact for the hundreds of millions of people who use ChatGPT every day.
Dreaming V3 is an upgraded memory architecture for ChatGPT, launched on June 4, 2026. It improves how the chatbot retains context, maintains conversational continuity, and keeps past references relevant to current conversations.
But describing it as a memory upgrade undersells what actually changed. Dreaming V3 makes memory automatic, not manual. A background process synthesizes ChatGPT's memory from many conversations without explicit "remember this" requests, and it replaces the saved-memories list as the standalone foundation rather than supplementing it.
The old system required you to explicitly tell ChatGPT to remember something. Dreaming V3 runs continuously in the background after each conversation ends, reading across your entire history and synthesizing what matters: your preferences, your ongoing projects, your working style, your constraints, and your time-sensitive context. It updates without you touching it.
Three specific improvements that matter:
Factual recall in OpenAI's internal evaluations reached 82.8% with the 2026 architecture, up from 67.9% in the 2025 system. The accuracy of maintaining user settings such as "answer briefly" or "do not use honorifics" across multiple sessions reached 71.3%. The most innovative feature is memory self-update: a memory like "I have a business trip to Singapore in July" is automatically rewritten to "I went to Singapore" after the trip. The accuracy with which the AI recognizes the passage of time and updates memories without user intervention is 75.1%.
That time-sensitive updating is the genuinely new capability. Previous memory systems stored facts statically. Dreaming V3 understands that facts change over time and revises them accordingly without you having to intervene.
The rollout began on June 4, 2026, with ChatGPT Plus and Pro subscribers in the United States getting first access. OpenAI has confirmed plans to extend Dreaming V3 to additional subscription tiers and international markets, and free-tier access is reportedly in the works, although no specific timeline has been confirmed.
The honest concern: A memory that updates itself while you are away is more useful, and harder to fully inspect, than one you wrote yourself. That trade-off, convenience for legibility, is about to be tested on hundreds of millions of users. OpenAI gives you the ability to view, edit, and delete any stored memory, but the default experience is that ChatGPT is learning about you continuously, and you may not always know exactly what it has concluded.
The rollout will not be seamless. Server capacity limits mean Free tier users in some regions will not get Dreaming V3 until late July 2026. Early adopters on Reddit have reported occasional memory conflicts where contradictory preferences cause the system to ask for clarification more often than expected. OpenAI acknowledges these bugs.
What competitors are doing: Google's Gemini Advanced has been testing its own persistent memory model, internally called "Evergreen," since March 2026. Apple's on-device AI for iOS and macOS, dubbed "Remembrance," focuses on privacy-first local memory, albeit with limited cross-app intelligence.
Memory is now a competitive front that every major AI platform is fighting on simultaneously. The difference in approach is meaningful: ChatGPT's Dreaming V3 is cloud-synthesized and automatic. Apple's Remembrance is local and private. Google's Evergreen is somewhere in between. Which approach users prefer will tell the industry something important about how much convenience people will trade for privacy.
2. Microsoft Built Seven AI Models in Secret and Just Revealed Them All
Microsoft used Build 2026 on June 2 to unveil seven new in-house MAI models across reasoning, coding, image, voice, and transcription. The timing and the scope of the announcement tell you more than the technical details do.
Microsoft has been OpenAI's most important partner and largest investor for years. Every flagship Microsoft AI product runs on OpenAI models. Build 2026 is the moment Microsoft told the world it is done depending entirely on one partner's roadmap.
The seven models, what they do, and where they are live:
MAI-Thinking-1 is Microsoft's first reasoning model, a 35B active parameter mixture-of-experts architecture with a 256K context window. It competes in the medium-sized weight class on reasoning and software engineering tasks.
MAI-Code-1, the inference efficient coding model tuned for GitHub, is now available in Copilot and VS Code. MAI-Transcribe-1.5 combines state-of-the-art accuracy across 43 languages. MAI-Voice-2 is available in more than 15 additional languages with new voice options. MAI-Image-2-Efficient is rolling out to Copilot and Bing.
MAI-Image-2.5 brings what Microsoft says is a step-change in image editing quality. These models are going beyond Microsoft's own platforms and will be available on Fireworks AI, Baseten, and OpenRouter.
The availability on Fireworks AI, Baseten, and OpenRouter is the strategic signal: Microsoft is not just building models for its own products. It is building model credibility in the developer market, the same market where OpenAI's API and Anthropic's Claude compete every day.
Also announced at Build 2026:
GitHub Copilot switches all plans to token-based AI Credits billing on June 1, to heavy developer backlash. GitHub ships the agent-native Copilot desktop app in technical preview and makes the Copilot SDK generally available across Node.js, Python, Go, .NET, Rust, and Java on June 2. The Majorana 2 quantum chip was announced, and Project Solara targets AI agent devices.
Microsoft IQ is now generally available across GitHub Copilot, Foundry, and Copilot Studios. It is the new context layer that feeds AI agents real workplace knowledge through Work IQ from M365 Signals, structured business data through Fabric IQ, and fast web grounding through the new Web IQ. Frontier Tuning, available in private preview, allows agents to learn how your business operates within your compliance boundaries.
The GitHub Copilot billing switch deserves its own callout. Token-based AI Credits billing went live June 1, meaning developers who use Copilot heavily for agentic and multi-step tasks will now see variable monthly bills instead of a flat subscription. The developer community reaction was swift and negative. This is a genuine cost concern for teams running automated workflows through Copilot at scale, and it is worth auditing your token usage now before the first variable bill arrives.
The honest read on Build 2026: The most important enterprise impact may be cost routing, governance, and model choice rather than a visible change in the Copilot interface. Microsoft is clearly trying to build its own model credibility, separate from the OpenAI relationship. Seven models across five categories, all live or in preview, is a meaningful signal of Microsoft's intent. Whether these models match GPT-5.4 or Claude Opus 4.8 in capability is a separate question from whether Microsoft needed to ship them for strategic reasons. It did.
3. The US Federal AI Bill That Could Override Every State Law
This is the week's most consequential story for the long-term regulatory landscape, and it received less coverage than it deserved.
The headline provision: a three-year preemption of state AI laws related to the development of frontier AI models. In a 269-page bill, Congress proposed effectively freezing every state-level AI regulation that has been passed or is being considered, for at least three years.
This matters because a significant number of meaningful AI protections currently exist at the state level. Colorado's AI Act, for example, includes anti-discrimination requirements for AI systems used in consequential decisions. New York, California, and Illinois have passed or are considering similar legislation. The federal bill's preemption clause would override all of it.
Preempting state laws without matching federal protections is a gamble. The Colorado AI Act, which this would freeze, includes actual anti-discrimination requirements. The federal bill's "general applicability" carve-out for states is vague enough to litigate for a decade.
The AI industry, broadly, supports federal preemption because it simplifies compliance. Navigating 50 different state AI laws is genuinely difficult for companies building national or global products. The question is whether the federal bill creates meaningful protections to replace the state ones it would freeze, or simply removes the protections without replacing them.
This bill will generate significant legal and policy debate over the coming months. If it passes in its current form, it will be the most significant AI governance action in US history and will shape the regulatory environment for AI development for at least the rest of the decade.
4. xAI Won a US Government Contract and OpenAI Updated Its Science Model
Two stories from the week that both signal where enterprise AI spending is going.
xAI wins the federal government:
The US government gave xAI a sweeping federal AI contract. Combined with Grok Build, this week represents xAI's most significant product push since the company was created. From SpaceX's acquisition of xAI in February 2026 through this week, xAI has gone from a model lab to a full enterprise AI stack with government, developer, and consumer distribution channels.
Grok entering the US federal government market is a significant competitive development. OpenAI, Microsoft, and Google have all been building government AI practices for years. xAI landing a federal contract within months of its SpaceX acquisition suggests the combined entity is moving faster on enterprise and government sales than anyone anticipated.
OpenAI updates GPT-Rosalind for life sciences:
On June 3, OpenAI updated GPT-Rosalind, its model built for life sciences research. The update folds in GPT-5.5's agentic coding and tool use and adds two Codex plugins, Life Sciences Research and Life Sciences NGS Analysis, that connect the model to more than 50 scientific databases. OpenAI says the model leads GPT-5.5, Grok 4.3, and Gemini 3.1 Pro on its LifeSciBench evaluation while using 31 percent fewer tokens than GPT-5.5 on genomics workloads. The model is in research preview to eligible organizations worldwide through a trusted-access program.
A model that leads the field on life sciences benchmarks while using 31% fewer tokens than its predecessor is the kind of efficiency gain that makes AI genuinely affordable for research institutions that previously could not justify the compute costs. Access to 50-plus scientific databases through integrated Codex plugins means GPT-Rosalind can run a literature review, analyze genomic sequences, and draft a research summary in a single session that previously required three separate tools.
5. Claude Opus 4.1 Is Being Deprecated
On June 5, Anthropic notified developers that Claude Opus 4.1 is deprecated, with retirement on the Claude API scheduled for August 5, 2026.
This is a practical action item for any developer or team still running production workloads on Opus 4.1. You have until August 5 to migrate to Opus 4.7 or 4.8. The migration is straightforward but should be planned now, not in late July.
The deprecation timeline is consistent with Anthropic's approach of maintaining two to three active generations before retiring older models. With Opus 4.8 now the flagship, 4.1 being retired is expected. The August 5 date gives teams approximately two months to update their API calls and test the new model in staging before it matters in production.
6. The Real Market Share Data: The Single-Platform Era Is Over
This week, the June 2026 edition of the generative AI chatbot market share report landed, and the numbers tell the clearest story yet about where the competitive landscape actually stands.
ChatGPT remains the leader at 54.7% of worldwide web visits across the seven largest AI chatbots, down from 76.5% in February 2025. Google Gemini is second at 27.4%, up roughly 104% in six months, making it the fastest-scaling large assistant by web traffic. Claude's numbers are the most striking: 8.2% worldwide web-visit share, but growing 306% in a single quarter, from 203 million web visits in January 2026 to 824 million in April 2026. In the United States specifically, Claude's web-visit share is 12.5%, ahead of its global average. DeepSeek holds 4.1% and Grok 2.8% of global web visits.
Let's unpack what these numbers actually mean, because the raw percentages can be misleading.
ChatGPT losing 22 points of market share in 15 months is a dramatic number. But it lost share in a market that grew from roughly 2 billion monthly visits to over 10 billion. ChatGPT's absolute user base is larger than it has ever been. What happened is that the market expanded so fast that other platforms captured the growth without requiring ChatGPT to shrink in absolute terms.
Claude's 306% quarterly growth is the number that should be generating the most discussion and is not. Anthropic wins approximately 70% of head-to-head enterprise deals against OpenAI among new business purchasers. Claude generates the highest average session time of any major AI platform at 34.7 minutes per daily user. Session time is the metric that matters most for product quality, because it measures whether people are actually getting value from the tool or bouncing. 34.7 minutes per daily user is significantly higher than any competitor.
The market is two-tiered. The top two platforms, ChatGPT and Gemini, control the majority of web traffic, with the next competitors sharing a much smaller portion. The seven apps drew a combined 10.07 billion web visits in April 2026.
One important caveat on the Gemini numbers: Gemini is pre-installed on Android and Workspace, which inflates "web traffic" share for users who did not actively choose it. The Cloudflare ranking, which reflects what users actively go to, tells a different story and shows ChatGPT maintaining a stronger lead on actively initiated visits.
The honest summary of the market in June 2026: ChatGPT leads on volume, Claude leads on engagement and enterprise win rates, Gemini leads on distribution-driven growth, and xAI and DeepSeek are larger factors than most coverage suggests.
Everything That Happened: Quick Reference Table
| Story | Date | Status | Who It Affects |
|---|---|---|---|
| ChatGPT Dreaming V3 | June 4 | Live for Plus and Pro US | All ChatGPT users |
| Microsoft Build 2026 MAI models | June 2 | Live, MAI-Code-1 in Copilot and VS Code | Developers, GitHub Copilot users |
| GitHub Copilot token billing switch | June 1 | Live now | All GitHub Copilot subscribers |
| US Federal AI Bill 269-page preemption | June 1 | In Congress | AI industry, state regulation watchers |
| xAI Grok federal government contract | June 7 | Live | US government, enterprise AI market |
| OpenAI GPT-Rosalind life sciences update | June 3 | Research preview | Life sciences researchers, biopharma |
| Claude Opus 4.1 deprecation notice | June 5 | Retiring August 5, 2026 | Developers on Opus 4.1 API |
| AI chatbot market share report | June 8 | Published | Everyone choosing or building AI tools |
What This Week Actually Means
Three structural shifts are visible in this week's news, taken together.
The first is that memory is becoming infrastructure. ChatGPT's Dreaming V3 is not a feature, it is an architectural bet that persistent, automatic, self-updating memory will become the baseline expectation for AI assistants. Every major platform is building a version of this: Gemini Evergreen, Apple Remembrance, Claude's own memory system. Within 12 months, the question will not be whether your AI assistant remembers you, but how it does so and who controls what it stores.
The second is that the platform consolidation era in AI is ending faster than analysts predicted. The AI chatbot market grew from 2 billion to 10 billion monthly visits in 15 months. ChatGPT still leads with 54.7%, but 10 credible competitors are taking real share. Microsoft is building its own models. xAI is in the federal government. OpenAI is going public. The days when one platform defined the category are behind us.
The third is that the regulatory window is narrowing. A 269-page federal bill proposing to freeze all state AI laws for three years is not the end of the regulatory conversation, it is the beginning of a new and more consequential phase. If it passes, the rules of AI development in the United States will be set at the federal level for at least the rest of the decade, and the period where states could experiment with different approaches will be over.
All three of these dynamics will accelerate, not slow down, in the second half of 2026.
Frequently Asked Questions
What is ChatGPT Dreaming V3?
Dreaming V3 is a new ChatGPT memory architecture that OpenAI launched on June 4, 2026. Unlike the previous system that required users to manually save memories, Dreaming V3 runs a background process after every conversation and automatically synthesizes what ChatGPT should remember about you, including your preferences, ongoing projects, working style, and time-sensitive context. It updates memories as facts change over time, a feature called memory self-update. Internal evaluations showed factual recall improved from 67.9% to 82.8%. The rollout started with Plus and Pro subscribers in the US.
What did Microsoft announce at Build 2026?
At Build 2026 on June 2, 2026 in San Francisco, Microsoft unveiled seven new in-house MAI models: MAI-Thinking-1 (its first reasoning model, a 35B MoE with a 256K context window), MAI-Code-1 (a coding model now live in GitHub Copilot and VS Code), MAI-Code-1-Flash, MAI-Image-2.5 (image editing), MAI-Transcribe-1.5 (speech-to-text in 43 languages), MAI-Voice-2 (speech synthesis in 15-plus languages), and MAI-Image-2-Efficient. Microsoft also announced the GitHub Copilot desktop app in technical preview, Microsoft IQ, Project Solara, the Majorana 2 quantum chip, and switched GitHub Copilot to token-based billing.
What is MAI-Code-1 and is it available?
MAI-Code-1 is Microsoft's inference-efficient coding model, specifically tuned for GitHub and announced at Build 2026 on June 2. It is live now in GitHub Copilot and VS Code. It will also be available on Fireworks AI, Baseten, and OpenRouter.
What is the US federal AI bill from June 2026?
The US federal AI bill, introduced in early June 2026, is a 269-page piece of legislation that includes a three-year preemption of state AI laws related to frontier model development. It would override existing state AI legislation including the Colorado AI Act, which includes anti-discrimination requirements. The bill has not yet passed and is generating significant debate about whether federal preemption without equivalent federal protections removes consumer safeguards without replacing them.
What is the AI chatbot market share in June 2026?
According to the June 2026 Momentic market share report using Similarweb data, ChatGPT leads with 54.7% of worldwide web visits among the seven largest AI chatbots, down from 76.5% in February 2025. Google Gemini is second at 27.4%, up roughly 104% in six months. Claude has 8.2% worldwide web-visit share but grew 306% in a single quarter. DeepSeek holds 4.1% and Grok 2.8%. In the United States, ChatGPT leads at 58.9%, Gemini at 19.2%, and Claude at 12.5%.
What happened to Claude Opus 4.1?
Anthropic issued a deprecation notice for Claude Opus 4.1 on June 5, 2026. The model will be retired on the Claude API on August 5, 2026. Developers and teams running production workloads on Opus 4.1 should migrate to Claude Opus 4.7 or 4.8 before that date.
What did OpenAI announce for GPT-Rosalind?
OpenAI updated GPT-Rosalind on June 3, 2026. The update adds GPT-5.5's agentic coding and tool-use capabilities and integrates two new Codex plugins: Life Sciences Research and Life Sciences NGS Analysis, connecting the model to over 50 scientific databases including genomics, proteomics, spatial transcriptomics, medicinal chemistry, and applied genetics databases. It uses 31% fewer tokens than GPT-5.5 on genomics workloads and is available in research preview to eligible organizations through a trusted-access program.
Final Thoughts
The second week of June 2026 looked nothing like a normal product cycle. ChatGPT got a new brain. Microsoft declared architectural independence from OpenAI. Congress proposed the most significant AI governance action in US history. xAI landed in the federal government. Real market data confirmed that the single-platform era in AI is over.
The pattern running through all of it is speed. The iteration cycle has compressed to weeks. The regulatory response is accelerating to match. The competitive dynamics that were theoretical in 2025 are now visible in market share numbers, deprecation notices, and 269-page bills.
If you are building on AI, you need to audit your tool stack now, before the variable billing and the deprecation timelines catch you in production. If you are using AI, the tools available to you today are meaningfully better and more personalized than they were 90 days ago.
The third week of June brings the World Cup kickoff, the expected OpenAI IPO filing going public, and Gemini 3.5 Pro in its confirmed June release window. The pace is not slowing down.
Published June 9, 2026. Sources: OpenAI official blog, Microsoft official Build 2026 blog, Momentic/Similarweb market share report, TechTimes, Windows Forum, Dev Weekly by Ajit Singh, Tom's Guide, Gizbot, Build Fast With AI daily briefings June 2 through 8.
Related: AI News First Week June 2026 · Every Major AI Announcement at Google I/O 2026 · Is AI Really Taking Your Job? What the 2026 Data Says
Tags: ChatGPT Dreaming V3, ChatGPT memory update June 2026, Microsoft Build 2026 MAI models, MAI-Code-1 GitHub Copilot, US federal AI bill 2026, xAI Grok government contract, GPT-Rosalind life sciences update, Claude Opus 4.1 deprecated, AI chatbot market share June 2026, AI news second week June 2026, GitHub Copilot token billing
