JadidJust-in AI Developments, Incidents & Deadlines
June 2026
SMTWTFS123456789101112131415161718192021222324252627282930

AI news, Saturday, June 6, 2026

11 stories

TLDR

  • Gemini Interactions API legacy schema dies June 8 — Two-day deadline: upgrade Python/JS Gemini SDK to ≥2.0.0 or automated pipelines break this weekend.
  • VS Code 1.123 (Jun 3) — Agent session sync across machines, 1M-token context, /chronicle for natural-language session recall; update now.
  • Microsoft MAI models (Jun 2) — 7 in-house AI models at Build 2026; MAI-Thinking-1 (35B) matches Opus 4.6 on SWE-BenchA benchmark that tests whether a model can resolve real GitHub issues by editing code in popular Python repositories.Jimenez et al. 2023 ↗ Pro; MAI-Code-1-Flash (5B) hits 51% SWE-Bench Pro at a fraction of the cost.
  • Trump signs AI Executive Order (Jun 2) — Voluntary 30-day pre-release model review for national security risk + AI cybersecurity clearinghouse; narrowed from an earlier 90-day mandatory draft.
  • MiniMax M3 API live (Jun 1) — First open-weightA model whose trained weights are published for anyone to download and run.NTIA ↗ model claiming frontier coding (59% SWE-Bench Pro) + 1M-token context + native multimodal computer use; weights expected ~Jun 10–11.

Action Items

Things your engineering team should look at or do this week.

  • Update Gemini SDK before June 8 — The legacy response schema (outputs array, response_mime_type) is permanently removed Sunday. Upgrade the Python or JS Gemini SDK to ≥2.0.0 (auto-opts into the new steps/response_format schema). Any Gemini API pipeline not updated breaks this weekend. (Gemini API docs | AI Certs)

  • Claim Anthropic Agent SDK credits before June 15 — Starting June 15, Agent SDK / claude -p / Claude Code GitHub Actions integration moves to a separate monthly credit pool ($20–$200/mo by plan, billed at API list rates). Overflow is off by default — claim your credits now and configure overflow preference, or automated workflows will silently stop at the ceiling mid-month. (TechTimes | The New Stack)

  • Evaluate MAI-Code-1-Flash for high-volume agent pipelines — At 5B parameters with 51% SWE-Bench Pro, it offers dramatically lower inferenceRunning a trained model on new input to get an output, as opposed to training it.NVIDIA ↗ cost than frontier models for code-gen tasks. Available on Azure AI Foundry, Fireworks, Baseten, and OpenRouter now. (CNBC)

Models & Tools

  • MiniMax M3 (Jun 1) Open Source — First open-weight model combining frontier-tier coding (59.0% SWE-Bench Pro, exceeding GPT-5.5 and Gemini 3.1 Pro), a 1M-token context windowHow much text a model can see and use at once while writing a reply.Anthropic ↗, and native multimodal computer use. API is live now; model weights and technical report releasing ~Jun 10–11 on Hugging Face and GitHub. Note: benchmarks are self-reported; independent validation pending. (MiniMax Blog | TechTimes on unverified benchmarks)

  • Microsoft MAI Family (Jun 2) Infra — 7 in-house models launched at Build 2026, trained without OpenAI distillationTraining a smaller model to copy a larger model's outputs, so it is cheaper to run.Hinton et al. 2015 ↗. Key entries: MAI-Thinking-1 (35B, 256K context, matches Opus 4.6 on SWE-Bench Pro coding), MAI-Code-1-Flash (5B, 51% SWE-Bench Pro — rivals models 7× its size on coding tasks), MAI-Transcribe-1.5 (43 languages), MAI-Voice-2 (15 languages, fine-grained emotional control). All available on Azure AI Foundry plus Fireworks, Baseten, and OpenRouter. (CNBC | Thurrott | Microsoft AI Blog)

  • VS Code 1.123 (Jun 3) Agents — Key agenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗ upgrades: cross-machine agent session sync via GitHub account (chat history, touched files, repo context), 1M-token context support for Opus 4.7 and GPT-5.5, new /chronicle command to query past sessions in natural language and auto-generate standup notes, and parallel side-by-side agent sessions. (Release notes | VS Magazine)

  • GPT-Rosalind update (Jun 3–4) — OpenAI cut genomics compute costs and expanded global access to its biology/drug-discovery reasoning modelA model that works through hidden reasoning steps before its final answer. Those steps are billed as output.OpenAI ↗ (still in Trusted Access, not public GA). Also launched the Rosalind Biodefense program, expanding access to national security researchers. (TechTimes | OpenAI)

APIs & Pricing

  • Google: Gemini Interactions API legacy schema removed June 8 — Breaking: outputs array replaced by steps, response_mime_type replaced by polymorphic response_format. Affects any code using v1beta Interactions API directly. Upgrading to Python ≥2.0.0 or JS ≥2.0.0 SDK handles the migration automatically. (API docs | Changelog)

  • Google: Gemini CLI deprecated June 18 — The original Gemini CLI retires June 18; Google is pushing migration to the new Antigravity CLI. If your team runs Gemini CLI in scripts or CI, migrate before then. (KuCoin News | AI Builder Club)

  • Anthropic: Agent SDK / Claude Code billing split June 15 Agents — Programmatic Claude (Agent SDK, claude -p, GitHub Actions CI) moves to a separate monthly credit pool ($20 Pro / $100 Max 5× / $200 Max 20×) billed at API list rates. Interactive chat, Claude Code terminal, and Cowork are unaffected. No auto-overflow, no rollover. (TechTimes | Coder's Era)

AI in Production

  • Claude global API outage (Jun 2) — All Anthropic API endpoints (Opus 4.6 through Sonnet 4.6) plus the Claude Code CLI showed elevated error rates in a global service disruption; fix deployed by 10:42 UTC. The incident is prompting fresh discussion about AI resilience as agent workflows become critical infrastructure. (Thoughtworks write-up)

  • Azure East US power failure takes down Copilot (Jun 1) Infra — A thunderstorm knocked out primary and backup power at an Azure East US data center, causing widespread Copilot degradation starting May 29. Azure VP Mark Russinovich's RCA cited "a rare synchronization failure in the primary and backup power control systems." Takeaway: AI-dependent workflows need multi-region fallback, not just retry logic. (Windows News)

Industry & Business

  • AI-attributed layoffs hit ~80K workers in 2026 (Jun 2026) Hiring — 212 layoff events YTD have cut 134K workers; roughly 17% are explicitly attributed to AI. Notable: Intuit cut 3,000 (17% of workforce), Meta began cutting 8,000, Cisco cutting thousands — all citing AI investment realignment. Paradox: companies simultaneously report inability to hire enough senior AI/infra talent, compressing junior-heavy workforce pyramids. (CBS News | AllWork.Space)

Policy & Safety

  • Trump signs AI Executive Order (Jun 2) Policy — The "Promoting Advanced AI Innovation and Security" EO establishes: (1) voluntary 30-day pre-release review of frontier models by federal agencies for national security risk (trimmed from an earlier 90-day mandatory proposal after industry pushback); (2) an AI cybersecurity clearinghouse for sharing vulnerability data; (3) directives for agencies to build AI capability benchmarks. Labs keep control over release timing; participation is not mandatory. (TechCrunch | NPR | CNBC)

Type to search · ↑↓ to move · enter to open