Thursday, June 4, 2026

TLDR

Anthropic splits Agent SDK / Claude Code usage into a separate credit pool on June 15 — programmatic workflows get their own dollar-denominated credits; claim window opens ~June 8, deadline June 15.
DeepSeek nears $7.4B first-ever funding at $52–59B valuation (Jun 3) — Tencent, CATL, state AI funds, and founder Wenfeng Liang reportedly participating.
GitHub Copilot issues $30–$70 promo credits to cushion tokenA small unit of data, such as part of a word, that AI models read and generate. Many AI services price by tokens in and out.NVIDIA ↗ billing shock (Jun 3) — no structural rollback; credits deplete in hours for agent-heavy workflows.
NVIDIA Cosmos 3 launches at COMPUTEX (Jun 1) — open physical AI world model for robotics/AV; Super + Nano available now, Edge coming soon.
AlphaSense closes $350M at $7.5B (Jun 3) — enterprise intelligence at $600M+ ARR, nearly doubling previous valuation.

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

Claim your Anthropic Agent SDK credit before June 15 — Programmatic calls via Agent SDK, claude -p, Claude Code GitHub Actions, and third-party apps authenticating through the SDK will no longer draw from subscription usage limits starting June 15. New separate pool: $20/mo (Pro), $100/mo (Max 5x), $200/mo (Max 20x) billed at standard API rates. A claim email arrives ~June 8; you must actively opt in. Audit your programmatic usage this week and decide whether to bump subscription tier or enable overflow billing caps before the deadline. (InfoWorld | Codersera)
Set a GitHub Copilot spending limit now — Token-based billing is live since June 1; teams running agenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗ or chat-heavy Copilot workflows are reporting monthly credits depleted within hours of each billing cycle reset. There are no automatic guardrails — set a hard spending cap in settings today. (GitHub Community | gHacks)

NVIDIA Cosmos 3 (Jun 1) Open Source Robotics/IoT — Open world foundation model for physical AI on a mixture-of-transformers architecture: native vision reasoning, world simulation, and action generation in a single model for synthetic training data generation and robot policy development. Cosmos 3 Super (high physics fidelity) and Nano (sub-second inferenceRunning a trained model on new input to get an output, as opposed to training it.NVIDIA ↗) available now; Edge variant for on-device deployment coming soon. (NVIDIA Newsroom | Axios)
Microsoft Execution Containers (MXC) — preview (Jun 2) Infra Agents — New lightweight sandboxing primitive from Azure: define security policies once, and any agent runtime can spin up a compliant container automatically without custom Dockerfile work. Addresses the long-standing "how do you safely run agent-generated code in production?" problem. (Microsoft Build)
Microsoft Work IQ APIs — GA June 16 (Jun 3) Agents — Enterprise context layer for agents (organizational graph, email summaries, SharePoint content) opens to third-party developers on June 16, one week after Build. Pairs with Web IQ (external web grounding, already GA) and Microsoft IQ (Azure infra context) announced earlier this week. (Microsoft Build)

Anthropic: Agent SDK usage splits to separate credit pool on June 15 (May 14 announced, Jun 15 deadline) — Interactive Claude use — chat, Claude Code run manually in terminal, Claude Cowork — is unaffected. Only automated/programmatic workflows move to the new pool. Heavy Opus-based pipelines or parallel CI runs will exhaust the credit in days at standard API pricing; prior subscriptions subsidized agent usage at ~15–30× below API rates. Claim your credits when the email arrives (~Jun 8) and configure overflow billing before June 15. (InfoWorld | devtoolpicks)
UPDATE — GitHub Copilot: promotional credits offered, billing model unchanged (Jun 3) Trending — After widespread developer backlash over token-based billing (some teams projecting 10–50× cost increases), GitHub's response is temporary promotional credits — +$30/mo for Business, +$70/mo for Enterprise — during June, July, and August 2026. No structural rollback to flat-rate pricing has been announced; GitHub is directing users to spending limits and usage dashboards as the primary cost controls. Code completions and Next Edit Suggestions remain unlimited and unmetered. (TechCrunch | gHacks)

DeepSeek nears $7.4B first-ever funding at $52–59B valuation (Jun 3–4) Open Source — Chinese open-weightA model whose trained weights are published for anyone to download and run.NTIA ↗ AI lab reportedly close to closing its first external round; Tencent (~~¥10B), CATL (~~¥5B), NetEase, JD.com, Hong Kong-based IDG Capital and Monolith Capital, state-backed AI funds, and founder Wenfeng Liang (¥20B) are named as prospective investors. Round could close within weeks; valuation range ¥350–400B ($52–59B). No disclosed model roadmap impact yet. (Bloomberg | TechNode)
AlphaSense closes $350M at $7.5B valuation (Jun 3) — Enterprise AI market intelligence platform (AI search over earnings, filings, research, expert transcripts) nearly doubles its previous $4B valuation; $600M+ ARR confirmed as of Q1 2026 and total funding over $1B. Round led by Vitruvian Partners, Accenture Ventures, and J.P. Morgan Asset Management, with CapitalG, Goldman Sachs Alternatives, D.E. Shaw Ventures, and Viking Global also participating. (Tech Startups)

GitHub Copilot agent workflows depleting credits within hours (Jun 1–3) Trending — Teams with agentic Copilot usage (multi-turn chat, code review, CI/CD integrations) report consuming entire monthly credit allocations on day one of the billing cycle. Root cause: token costs for agent loops with long context windowsHow much text a model can see and use at once while writing a reply.Anthropic ↗ are dramatically higher than prior flat-rate assumptions; one user reported projected costs ballooning from ~$29/mo to ~$750/mo. Lesson: profile per-workflow token consumption before setting hard spending caps — completions remain free and unmetered, but chat/agent calls are not. (gHacks | how2shout)

OpenClaw Open Source Agents — Personal AI gateway that connects local models (via Ollama) to 50+ messaging and productivity integrations (WhatsApp, Slack, Telegram, Discord, Signal, iMessage) and runs entirely on-device. Surged from 9K to over 210K GitHub stars since late January 2026 — the fastest-growing open-source AI project of 2026 by star velocity. Useful reference architecture for teams building private local agents with MCP-style tool integrations. (GitHub)