--- title: Apple rebuilds Siri on Gemini at WWDC, Xiaomi MiMo hits... description: Apple's WWDC 2026 featured a full Siri rebuild on a licensed Google Gemini model (~1.2T parameters), launching as a standalone app with a three-tier privacy... canonical: https://daily.dev/posts/apple-rebuilds-siri-on-gemini-at-wwdc-xiaomi-mimo-hits-1000-tokens-per-second-2uowxru1g twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Apple rebuilds Siri on Gemini at WWDC, Xiaomi MiMo hits 1000 tokens per second | daily.dev og:description: Apple's WWDC 2026 featured a full Siri rebuild on a licensed Google Gemini model (~1.2T parameters), launching as a standalone app with a three-tier privacy... og:url: https://daily.dev/posts/apple-rebuilds-siri-on-gemini-at-wwdc-xiaomi-mimo-hits-1000-tokens-per-second-2uowxru1g og:image: https://api.daily.dev/og/posts/2uOwXru1g.png og:image:alt: Apple rebuilds Siri on Gemini at WWDC, Xiaomi MiMo hits 1000 tokens per second og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Apple rebuilds Siri on Gemini at WWDC, Xiaomi MiMo hits 1000 tokens per second **[Agentic Digest](https://daily.dev/sources/agents_digest)** · 5 min read · 2 upvotes · 0 comments ## Summary Apple's WWDC 2026 featured a full Siri rebuild on a licensed Google Gemini model (~1.2T parameters), launching as a standalone app with a three-tier privacy architecture across on-device, Private Cloud Compute, and Google Cloud. iOS 27 opens third-party AI defaults, and Apple's Foundation Models framework goes open source. Xiaomi released MiMo-V2.5-Pro, a 1T open MoE model matching frontier coding benchmarks (78.9% on SWE-bench Verified) at significantly lower cost, with a new UltraSpeed mode exceeding 1,000 tokens/second on commodity hardware. OpenAI filed confidentially for an IPO, following Anthropic's similar move, with both racing to capture capital. A supply chain attack called Shai-Hulud compromised gpt-pilot and Microsoft repos, targeting AWS keys, GitHub secrets, and Claude Code hooks, with Microsoft disabling 70+ GitHub repositories after related Miasma malware was discovered. ## Content **TLDR:** Apple's WWDC 2026 keynote was dominated by a full Siri overhaul built on a licensed Google Gemini model, arriving as a standalone app with a three-tier privacy stack. Xiaomi's MiMo-V2.5-Pro is getting serious attention: a 1-trillion-parameter open MoE model that matches frontier coding benchmarks at a fraction of the cost, and a new UltraSpeed mode hitting 1,000+ tokens per second on commodity hardware. OpenAI filed confidentially for an IPO, following Anthropic's similar move, with both companies racing to capture capital before the window narrows. A supply chain attack dubbed Shai-Hulud hit gpt-pilot and Microsoft repos, targeting AWS keys, GitHub secrets, and Claude Code hooks. --- ## Apple rebuilds Siri on licensed Gemini, ships standalone app and three-tier privacy stack at WWDC Apple unveiled Siri AI at WWDC 2026, rebuilt around a custom Google Gemini model (~1.2 trillion parameters) licensed at roughly $1 billion per year. It arrives as a standalone app across iPhone, iPad, Mac, and Vision Pro, with personal context search, multi-step commands, and on-screen awareness. Privacy runs through three tiers: on-device for simple tasks, Private Cloud Compute for moderate requests, and Google Cloud for heavy reasoning — with Apple stating queries are processed statelessly and Google is barred from training on user data. iOS 27 also opens the door to third-party AI defaults including Claude and ChatGPT, and Tim Cook made his final WWDC appearance before handing the CEO role to John Ternus on September 1. The Foundation Models framework is going open source and now supports third-party providers, image input, and a new on-device vision model, with free Private Cloud Compute access for developers under 2 million first-time App Store downloads. [Read more](https://app.daily.dev/feed-by-ids?id=NkgqtzUP8&id=cYwVTYIfm&id=FVHk32jQP&id=RkwLGjaWU&id=3WHrlbqPG&id=ertp8mCSn&id=ay77d3AXV&id=pbWxaLaKq) ## Xiaomi MiMo-V2.5-Pro: 1T open MoE matches frontier coding benchmarks, UltraSpeed hits 1000 tokens/sec Xiaomi's MiMo-V2.5-Pro is a 1.02-trillion-parameter Mixture-of-Experts model with 42B active parameters, open-sourced under MIT. It scores 78.9% on SWE-bench Verified and 57.2% on SWE-bench Pro — competitive with Claude Opus 4.6 and GPT-5.5 — while using 40-60% fewer tokens per trajectory and priced at $1/$3 per million input/output tokens. The new UltraSpeed mode, developed with inference partner TileRT, breaks 1,000 tokens per second on a standard 8-GPU commodity node using selective FP4 quantization on MoE expert layers and DFlash speculative decoding, at 3x the standard price. For context, GPT-5.5 runs at roughly 68 tokens/sec and Claude Opus 4.6 at roughly 71. The FP4-DFlash checkpoint is open on Hugging Face, and a limited API trial runs June 9-23. [Read more](https://app.daily.dev/feed-by-ids?id=lPUDOHUoD&id=00a4CAIbk&id=srSKWsIjl&id=lL7ElOwRo&id=es74Jr8mk&id=oCsFZHFKB&id=aGUcAdf3e&id=dWVGjv5lz) ## OpenAI files confidentially for IPO as Anthropic race to capture capital OpenAI has filed confidentially for an IPO, following Anthropic's similar filing just over a week earlier. Anthropic is nearing its first quarterly profit and has seen 123% valuation appreciation year-to-date on secondary markets, versus OpenAI's 11.3%. OpenAI was last valued at $852 billion and projects burning $85 billion in 2028 even after doubling sales, with its CFO raising concerns about sustaining data center spending. Whoever IPOs first will set a valuation benchmark that constrains the other's pricing — and both are racing against a narrowing capital window. [Read more](https://app.daily.dev/posts/26Hvrl7YZ) ## Shai-Hulud supply chain attack hits gpt-pilot and Microsoft repos, targets Claude Code hooks A threat actor compromised the co-founder's GitHub account for gpt-pilot (33K stars) and force-pushed a 758KB obfuscated JavaScript payload targeting AWS keys, npm tokens, GitHub secrets, Kubernetes service accounts, and SSH keys. The malware plants persistence hooks specifically in Claude Code and VS Code, uses GitHub commit searches as a covert C2 channel, and abuses Sigstore to sign malicious npm packages with valid SLSA attestations. The attack was stopped twice by the ruff Python linter catching formatting violations in the injected code. Separately, Microsoft disabled at least 70 open-source GitHub repositories after the related Miasma malware was found in Azure and AI developer tool integrations including Claude Code and Gemini CLI. [Read more](https://app.daily.dev/feed-by-ids?id=NVzcFrIXe&id=F16SkZhxx) --- ## Also notable - **FrontierCode benchmark grades like a tech lead: Claude Opus 4.8 scores 13.4% on Diamond tier:** Cognition's FrontierCode benchmark, built with 40+ hours per task from open-source maintainers, evaluates code mergeability rather than just test passage — and even the best model, Claude Opus 4.8, scores only 13.4% on the hardest Diamond subset, with GPT-5.5 at 6.3% and Gemini 3.1 Pro at 4.7%. [Read more](https://app.daily.dev/feed-by-ids?id=cbekGWzxD&id=E1P7cCDhb) - **Vercel AI Gateway: DeepSeek token share surged to 17% in May while Anthropic grew spend share to 65%:** Vercel's June 2026 production index shows DeepSeek's token share jumped from under 1% to 17% following DeepSeek V4 Flash and V4 Pro launches priced 20-50x cheaper than Anthropic, yet its spend share stayed near 1%, while Anthropic grew spend share from 61% to 65% — teams are routing cheap volume tasks to DeepSeek and reserving frontier models for coding agents and back-office automation. [Read more](https://app.daily.dev/posts/CaIs1dY8H) - **Anthropic Computer Use agent: tasks completed 87% faster at 94% lower cost than search alone in Harvard study:** New Anthropic/Harvard research over three months found workers using the Computer Use autonomous agent finished tasks 87% faster and at 94% lower cost compared to search alone, with higher reported satisfaction. [Read more](https://app.daily.dev/posts/pReRl2VNL) - **AI productivity ceiling: Amdahl's Law analysis puts hard cap on shipping improvement at ~53%:** An MIT/Wharton paper analysis finds the parallelizable fraction of software delivery stays around 35% across all three generations of AI coding tools — autocomplete, sync agents, and async agents — putting the theoretical ceiling on total shipping improvement at roughly 53%, because planning, meetings, and alignment can't be parallelized. [Read more](https://app.daily.dev/posts/XmelmxBkm) - **Microsoft disabled 70+ GitHub repos after Miasma worm broke CI/CD pipelines globally:** Microsoft confirmed it notified a small number of affected customers after the Miasma supply chain attack disabled 73 Microsoft-owned GitHub repositories and broke CI/CD pipelines globally — the company's second such breach in weeks, with the first involving the Durable Task project in mid-May. [Read more](https://app.daily.dev/feed-by-ids?id=H3H6uJb7S&id=F16SkZhxx) ## Similar posts on daily.dev - [Xiaomi releases MIT‑licensed MiMo models for long‑running AI agents](https://daily.dev/posts/xiaomi-releases-mit-licensed-mimo-models-for-long-running-ai-agents-qgbigdop8) · InfoWorld · 2 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#security](https://daily.dev/tags/security), [#llm](https://daily.dev/tags/llm), [#openai](https://daily.dev/tags/openai) [View this post on daily.dev](https://daily.dev/posts/apple-rebuilds-siri-on-gemini-at-wwdc-xiaomi-mimo-hits-1000-tokens-per-second-2uowxru1g) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/apple-rebuilds-siri-on-gemini-at-wwdc-xiaomi-mimo-hits-1000-tokens-per-second-2uowxru1g","headline":"Apple rebuilds Siri on Gemini at WWDC, Xiaomi MiMo hits 1000 tokens per second","text":"Apple's WWDC 2026 featured a full Siri rebuild on a licensed Google Gemini model (~1.2T parameters), launching as a standalone app with a three-tier privacy architecture across on-device, Private Cloud Compute, and Google Cloud. iOS 27 opens third-party AI defaults, and Apple's Foundation Models framework goes open source. Xiaomi released MiMo-V2.5-Pro, a 1T open MoE model matching frontier coding benchmarks (78.9% on SWE-bench Verified) at significantly lower cost, with a new UltraSpeed mode exceeding 1,000 tokens/second on commodity hardware. OpenAI filed confidentially for an IPO, following Anthropic's similar move, with both racing to capture capital. A supply chain attack called Shai-Hulud compromised gpt-pilot and Microsoft repos, targeting AWS keys, GitHub secrets, and Claude Code hooks, with Microsoft disabling 70+ GitHub repositories after related Miasma malware was discovered.","url":"https://daily.dev/posts/apple-rebuilds-siri-on-gemini-at-wwdc-xiaomi-mimo-hits-1000-tokens-per-second-2uowxru1g","datePublished":"2026-06-09T04:18:51.038Z","dateModified":"2026-06-09T04:19:13.255Z","author":{"@type":"Organization","name":"Agentic Digest","logo":"https://media.daily.dev/image/upload/s--V91DY4ls--/f_auto,q_auto/v1772617267/logos/agents_digest","url":"https://daily.dev/sources/agents_digest"},"interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/agents_digest","name":"Agentic Digest"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Agentic Digest","item":"https://daily.dev/sources/agents_digest"},{"@type":"ListItem","position":3,"name":"Apple rebuilds Siri on Gemini at WWDC, Xiaomi MiMo hits 1000 tokens per second"}]} ```