Week 24Published

Apple flips into the AI platform war at WWDC + OpenAI files S-1 + US export controls cut Anthropic's Fable 5/Mythos 5 + Kimi K2.7-Code open-sourced + GLM 5.2 + multi-vendor alliances

applewwdcfoundation-modelscore-aianthropicopenaiipo-s1export-controlsmoonshotkimi-k2-7zhipuglm-5-2dxctcsoraclehitachiglasswingintelfoundrygemini
On this page

Date range: 2026-06-08 (Mon) → 2026-06-14 (Sun) Compiled: 2026-06-18 (full-week coverage — W24 data is closed) Sources: Apple Newsroom, Anthropic Newsroom, OpenAI Newsroom, GitHub Releases, Reuters / TradingView, CNBC, MacRumors, Fortune, Moonshot AI, Z.ai, QbitAI, Tom's Hardware, and more.

🌐 Lead story (full week 6/8-6/14)

"Apple becomes an AI platform overnight" + "OpenAI follows with an S-1 filing — the frontier IPO race begins" + "US export controls knock Fable 5 / Mythos 5 out of the global market" + "China's open-source June detonates — Kimi K2.7-Code + GLM 5.2 land the same week".

W24 is the highest-information-density week of 2026 so far —

  • 🍎 6/8 WWDC 2026 keynote: Apple Foundation Models 3 + AFM Cloud Pro publicly acknowledge joint training with Google Gemini; Core AI framework replaces Core ML, with an MCP-style plugin interface letting apps connect to Apple/Claude/GPT/Gemini at the same time; the new Siri ships in beta in July, iOS 27 ships in September; Cook's last WWDC.
  • 🟧 6/8 Hitachi joins Anthropic Project Glasswing + 6/11 DXC × Anthropic global alliance + 6/12 TCS × Anthropic global alliance — Anthropic signs three SI giants in regulated industries in a single week.
  • 🟩 6/10 OpenAI files a confidential S-1 draft — the frontier AI labs collectively enter the SEC process; the same week, 6/11 OpenAI × Oracle channel integration + 6/12 OpenAI acquires Ona (formerly Gitpod), disclosing Codex weekly actives >5M / +400% YoY; 6/14 OpenAI Academy adds three courses.
  • ⛔ 6/12 US export control order — Anthropic halts global access to Fable 5 / Mythos 5, locking out even Anthropic's foreign-national employees; Anthropic publicly states it disagrees with the cited jailbreak characterization.
  • 🟪 China's open-source June detonates: 6/12 Moonshot open-sources Kimi K2.7-Code (1T MoE / 32B active, coding +21.8% vs K2.6, $0.95/$4.00 per M) + 6/13 Z.ai ships GLM 5.2 (744B MoE / 1M context / dual reasoning tiers, MIT-licensed weights one week later).
  • 🔷 6/8 Intel Foundry wins Google AI chip mega-order (3M+ units by 2028, INTC +13%) + 6/11 Meta $115M Indianapolis data-center bootcamp + 6/12 Citi upgrades AMD to Buy ($575) + 6/12 xAI financial disclosure: $6.36B burned in 2025 / $2.4B burned in Q1 2026 on $7.7B capex.

⭐ Personal Top 5 (usable today / new this week)

Ranking by: real capability lift × low onboarding cost × direct cost / productivity gain

Rank Item One-liner
🥇 1 Moonshot open-sources Kimi K2.7-Code (6/12) 1T MoE / 32B active, Coding Bench v2 +21.8% / Program Bench +11.0%, 30% fewer reasoning tokens than K2.6; API priced $0.95 / $4.00 per M tokens (roughly 1/8 of Opus), Modified MIT license for commercial use. Personal coders should switch immediately from 6/12.20
🥈 2 Apple Foundation Models framework (6/8 WWDC) Swift developers get on-device 3B + server-side models, image input + custom skills + multimodal generation, with zero API fees + no rate limits + no telemetry. Personal iOS / macOS developers officially enter the "free AI capability" era.2
🥉 3 Z.ai GLM 5.2 (6/13) 744B MoE / 1M context, dual reasoning tiers High & Max; GLM Coding Plan available across all tiers (Lite/Pro/Max/Team), with MIT-licensed weights coming next week — the next shot from China's open-source camp.21
4 Core AI framework + MCP-style plugin interface (6/8) Personal apps can connect Apple's own models + Claude/GPT/Gemini simultaneously, managing multiple backends through a unified interface — the first standardized "cross-model" entry point for Apple ecosystem developers.
5 Claude Code v2.1.172 — sub-agents nest 5 levels + plugin marketplace search (6/10) Sub-agents can spawn sub-agents 5 levels deep; plugin marketplace gains search; long-session perf + background agent settings isolation get a bundled fix. Daily cc users should upgrade immediately.12

🏢 Team / Enterprise Top 5 (actionable this week)

Ranking by: strategic signal strength × deployment reach × security / compliance impact

Rank Item One-liner
🥇 1 OpenAI's strategic combo: S-1 draft (6/10) + acquires Ona (6/12) + Oracle channel (6/11) Three moves in one week — IPO prep + secure cloud sandbox internalized + enterprise procurement friction removed. Discloses Codex weekly actives >5M / +400% YoY. Gives every enterprise evaluating OpenAI's long-term path a complete picture: IPO countdown has started.13, Ona, Oracle)
🥈 2 Anthropic's regulated-industries leap: Hitachi (6/8) + DXC (6/11) + TCS (6/12) three-in-a-row Three global SI giants join Project Glasswing / Anthropic's Partner program in one week — TCS deploys Claude to 50K employees; DXC trains tens of thousands of Claude-certified forward-deployed engineers. Anthropic turns "embed enterprise AI into regulated-industry core systems" from concept into a buyable service. (DXC, TCS)
🥉 3 Apple Foundation Models + Core AI open to Swift developers (6/8) Gives every enterprise with an iOS / macOS app an "inline AI + zero cost + on-device privacy" inflection point; Apple overnight becomes a buyable AI vendor. Product decisions now need to consider an "Apple device-first AI" route.
4 US export controls — Anthropic halts Fable 5 / Mythos 5 (6/12) A blanket cut that even locks out Anthropic's foreign-national employees; multinational AI teams + foreign-employee access risk models need an immediate review. "AI export controls" has hit the inflection point from "military ChatGPT" to "frontier for everyone".18, Fortune)
5 Intel Foundry × Google AI chip order (6/8) + Citi upgrades AMD to Buy (6/12) Multi-fab strategy accelerates + AMD re-enters the frontier procurement shortlist — a hard signal to every GPU/TPU buyer that "you can no longer bet only on NVIDIA + TSMC".9, Citi)

🔵 Top two

Anthropic / Claude Code

Date Event One-liner
06-08 Hitachi joins Project Glasswing First Japanese infrastructure giant onboard; granted Claude Mythos Preview access; marks Glasswing's spread from US energy / industrials to Japan.8
06-10 Claude Code v2.1.172 — sub-agents nest 5 levels + Bedrock ~/.aws config + plugin marketplace search Long-session perf + background agent settings isolation + model picker / permission rules — a bundled fix.12
06-10 Claude Platform AWS Sandbox endpoint GET /v1/environments/{id}/work Pending-work listing endpoint for self-hosted sandboxes lands on Claude Platform on AWS, with companion IAM action GetEnvironment.11
06-11 Claude Code v2.1.173 Patches parsing of the 1M-context suffix in Fable 5 model IDs, plus the spurious "sandbox dependencies missing" warning on Windows.
06-11 DXC × Anthropic multi-year global alliance DXC trains tens of thousands of Claude-certified forward-deployed engineers, embedding Claude into core banking / aviation / insurance / manufacturing / government systems.14
06-12 Claude Code v2.1.174 wheelScrollAccelerationEnabled setting, /model Default family fix, Bedrock GovCloud region fix, VSCode usage attribution.
06-12 TCS × Anthropic global alliance TCS deploys Claude to 50K employees across 56 countries and is building Claude-driven products for finance / healthcare / public-sector clients.17
🔥 06-12 US export control order — Fable 5 / Mythos 5 global access cut at 5:21 pm ET The Department of Commerce issues a legally binding export control order prohibiting any foreign national (including Anthropic's foreign-national employees) from accessing Fable 5 / Mythos 5; Anthropic complies the same day but publicly states it disagrees with the cited jailbreak characterization.18, Fortune)

OpenAI / Codex

Date Event One-liner
06-08 Codex CLI v0.138.0-alpha.7 Another 0.138 alpha rolling build, no standalone changelog; stable tag remains at 0.137.0.
🔥 06-10 OpenAI files confidential S-1 draft with SEC OpenAI confidentially submits an S-1 draft, preserving optionality on a future IPO; timing undecided, disclosure forced early after anticipated leaks. The frontier AI labs collectively enter the SEC process.13
06-11 OpenAI × Oracle — Universal Credits unlocked Enterprise users' committed Oracle cloud spend can, within weeks, be directly applied to OpenAI model and Codex usage, bypassing fresh procurement / security review.15
06-11 Codex case study: University of Arizona black hole simulation The EHT collaboration uses Codex to help develop algorithms simulating the plasma around the M87 black hole, with the goal of producing the first-ever black hole video footage.16
🔥 06-12 OpenAI acquires Ona (formerly Gitpod) Codex gets "secure cloud sandbox infrastructure where agents keep working after disconnect"; discloses Codex weekly actives >5M / +400% YoY.19
06-14 OpenAI Academy adds three courses Built with BCG / Accenture / BBVA, giving teams a unified learning path: "understand AI → apply to daily work → orchestrate structured workflows with agents".23

🟢 Chinese camp

Moonshot / Kimi

  • 🔥 06-12 | Kimi K2.7-Code open-sourced — 1T MoE / 32B active, coding-specialized variant; Kimi Code Bench v2 +21.8% / Program Bench +11.0% / MLS Bench Lite +31.5%; 30% fewer reasoning tokens than K2.6; API $0.95 / $4.00 per M tokens (~1/8 the price of Claude Opus); ships with the Kimi Code terminal agent; Modified MIT license, commercial use allowed. Yang Zhilin's internal memo still pegs K3 for Q3 — K2.7-Code is the "coding puzzle piece" ahead of K3.20

Zhipu / GLM

  • 🔥 06-13 | Z.ai ships GLM 5.2 — 744B MoE / 1M context / dual reasoning tiers High & Max; GLM Coding Plan available across all tiers (Lite/Pro/Max/Team), with 1M-token context ~5x of 5.1; commits to MIT-licensed weights next week; no benchmark numbers disclosed at launch.21

DeepSeek

  • No major announcement this week — V4-Pro/V4-Flash (released 4/24) remains the latest; V4.1 multimodal + MCP is postponed, nothing landing before end of June.

🟠 Hardware & compute & power

NVIDIA

  • 06-10 | NVIDIA optimizes DiffusionGemma + Confidential Computing extends to Google Cloud — Google DeepMind's newly released DiffusionGemma (experimental diffusion text generation model) is optimized for RTX / RTX PRO / DGX Spark; NVIDIA Confidential Computing GPUs extend from Apple Private Cloud Compute to Google Cloud. (newsroom)
  • 06-12 | Blackwell posts first AgentPerf benchmark numbers + NVIDIA + MiniMax M3 long-context + GeForce NOW summer promo $70 off — Artificial Analysis launches AgentPerf, billed as the industry's first agentic-AI benchmark; developer.nvidia.com/blog publishes MiniMax M3 deployment guide. (blogs)

Intel / AMD

  • 06-08 | Intel Foundry wins Google AI chip contract (3M+ units by 2028) — first major customer for Intel's 18A node; INTC premarket +13%; Google diversifies TPU-class silicon supply away from Broadcom. (GuruFocus)
  • 06-12 | Citi upgrades AMD to Buy, price target $460 → $575 — cites Meta AI accelerator procurement commitment underestimated; Intel also rated overweight. (Yahoo Finance)

xAI / Tesla / Musk

  • 06-10 | xAI adds Public URLs for Files + Files API input storage + Gopuff ships Go (based on Grok) + ex-employee Devin Kim sues xAI/SpaceX alleging he was fired for raising Grok safety concerns — xAI hit with simultaneous product + legal news. (xAI news,27
  • 06-11 | eToro launches Tori AI investing agent (based on xAI models, embedding real-time market sentiment into the investing flow).
  • 06-12 | xAI 2025 financial disclosure: $6.36B annual burn / $2.4B Q1 2026 burn on $7.7B capex + Musk hints on X at an "upgraded" partnership with NVIDIA — xAI's financial pressure surfaces while Musk signals deeper NVIDIA ties. (Basenor analysis)

Other hardware / compute / power

  • 🔥 06-11 | Meta commits $115M to Indianapolis "America's Workforce Academy" — paid bootcamp (covering airfare / lodging / stipend), free tuition + job offer guarantee, targeted at training thousands of data-center + fiber-laying technicians. Pilot cities: Indianapolis, Baton Rouge, Houston, Columbus OH. A hyperscaler finally publicly admits that the data-center expansion bottleneck isn't just chips — it's people. (Indiana Capital Chronicle)
  • 06-10 | TSMC May revenue NT$416.98B (+1.5% MoM, +30.1% YoY) — monthly 6-K filed 6/10. (SEC)

🌍 Other vendor moves (aggregator view)

Apple — overnight becomes a "frontier AI platform"

  • 06-08 | Foundation Models framework gen-2 expansion + Core AI framework — see the lead-story section above; a "platform-open + multi-model routing" combo. (Apple Dev Docs)
  • 06-08 | Apple × Google Gemini joint training publicly acknowledged — Apple rarely admits external dependency; Gemini's reach spreads from Pixel to Apple. (Apple Newsroom)
  • 06-08 | "New Siri" July public beta + iOS 27 ships September — EU users locked out due to DMA. (CNBC live)
  • 06-08 | visionOS 27 Visual Intelligence + curved windows — Vision Pro's first true AI moment. (AppleInsider, Engadget)
  • 06-08 | macOS 27 "Golden Gate" + Cook's last WWDC keynote — signaling Apple's strategic succession period. (IndexBox)
  • 06-10/11/12 | WWDC26 dev sessions continue rolling out — Foundation Models sessions 241/242/319 cover framework open-sourcing, agentic API, Private Cloud Compute large models. (Apple Dev)

🧭 Trend analysis

1. The frontier AI labs collectively enter the IPO process — Anthropic (W23) + OpenAI (W24) file S-1 in the same month

5/28 Anthropic's Series H at a $965B valuation overtakes OpenAI; 6/1 Anthropic files S-1; 6/10 OpenAI files S-1 too. Both frontier labs enter SEC confidential filing within one month — the AI frontier IPO race is officially on. The two stories are different: Anthropic bets on "harness engineering + regulated-industry automation" (DXC + TCS + Hitachi three-in-a-row); OpenAI bets on "channel financialization + secure sandbox internalization + procurement friction removal" (Oracle + Ona + Partner Network). Capital markets will force these two paths to a verdict in H2 2026.

2. AI export controls move from "military ChatGPT" to "frontier for everyone" — Fable 5 shutdown is the watershed

Pre-W24 export controls were basically GPU + Hopper/Blackwell volumes; 6/12 the US Department of Commerce directly cuts Anthropic Fable 5 / Mythos 5 global access, locking out even Anthropic's foreign-national employees. This is the first case of "frontier AI models themselves placed under export control". Every multinational AI team's access risk model needs an immediate review: "can foreign-national employees access internal models?" "can offshore dev teams use frontier APIs?" "will G7 joint training trigger a block?". Anthropic states it disagrees with the cited jailbreak characterization but complies — law > disagreement.

3. Apple finally enters as an "AI platform" — cross-platform AI app dev needs re-evaluation

The real signal from Apple's WWDC is Foundation Models framework + Core AI MCP-style plugins — embedding AI capability into iOS / macOS apps shifts from "call cloud API" to "call local framework" + "device-side multi-model routing". Every AI startup with an iOS / macOS entry point needs to revisit "cost structure + privacy story + model routing architecture" over the next 12 months.

4. China's open-source June detonates — Kimi K2.7-Code + GLM 5.2 land the same week

6/12 Moonshot open-sources Kimi K2.7-Code (1T MoE / 32B active, coding +21.8%, $0.95/$4.00 per M) + 6/13 Z.ai ships GLM 5.2 (744B MoE / 1M context, MIT license). China AI's "June launch window" didn't miscarry — it just got pushed to mid-month. Both bet on "open source + low price + coding scenarios" — for the first time, small / mid teams have a real option to self-host frontier-grade coding agents. DeepSeek V4.1 still isn't out, but K2.7-Code + GLM 5.2 once again make the question concrete: "can China open-source hold its own against the Anthropic + OpenAI closed-source stack?"

5. Anthropic opens a second axis of "enterprises can't bet on just one vendor" in regulated industries

W22's Fujitsu signing both OAI + Anthropic the same day was multi-vendor on the procurement side; W24 Anthropic signing Hitachi + DXC + TCS in one week is multi-vendor on the SI channel side. TCS deploying Claude to 50K employees across 56 countries + DXC training tens of thousands of Claude-certified forward-deployed engineers — gives the industry a quantitative reference: "enterprise Claude deployment" shifts from PoC to SI-led project delivery.

6. Hyperscaler data-center expansion bottleneck shifts from "chips" to "people"

6/11 Meta commits $115M to the Indianapolis bootcamp to train data-center + fiber technicians — this is the first time in 2026 a hyperscaler has publicly admitted "the core bottleneck for data-center expansion is headcount of technicians". A "talent war" among hyperscalers will emerge over the next 2-3 years, introducing a new talent cost variable for any enterprise betting on hyperscaler-built AI compute.

📚 Further reading