The pre-clearance machine produced its first big result today: the Commerce Department signed off on a broad GPT-5.6 rollout after weeks of national-security testing, and OpenAI says the Sol/Terra/Luna trio goes broadly public Jul 9 — the first frontier model to clear the White House's new 30-day review and ship.
Elsewhere it's a housekeeping-and-money day: Fable 5 drops off subscription inclusion today (metered-only now), the OpenAI Assistants API has a hard Aug 26 shutdown worth putting on the calendar, Musk's AI lab completed its SpaceXAI rebrand as SPCX joined the Nasdaq-100, and legal-AI startup Norm hit a $1.2B unicorn valuation.
TLDR
- OpenAI clears the White House pre-deployment review — Commerce approves a broad GPT-5.6 (Sol/Terra/Luna) rollout, with public availability Jul 9 (Jul 8) Trending Policy
- Musk's xAI completes its rebrand to SpaceXAI (new logo, merged into SpaceX); SPCX joined the Nasdaq-100 on Jul 7 (Jul 6–7) Business
- Legal-AI startup Norm raises $120M Series C at a $1.2B valuation (Khosla-led), building an AI-native law firm (Jul 7) Business
- Housekeeping: Claude Fable 5 leaves subscription inclusion today — metered credits only ($10/$50 per M in/out); OpenAI Assistants API hard-shuts down Aug 26 (Jul 8 / deadline Aug 26) Pricing
Action Items
Things your engineering team should look at or do this week.
- Put the OpenAI Assistants API shutdown (Aug 26) on the migration calendar now. After that date every call to
/v1/assistants,/v1/threads, and/v1/threads/runsreturns an error — no grace period, and OpenAI won't ship an automated Threads→Conversations migration tool. Port POCs and prod bots to the Responses + Conversations APIs. (OpenAI migration guide) - Confirm Fable 5 billing before something breaks. As of today, Fable 5 is no longer included in Pro/Max/Team/select-Enterprise plans; access now requires usage credits ($10/M in, $50/M out). Any pipeline that assumed subscription-included Fable 5 either draws credits now or fails — verify credits are enabled or repoint to Opus 4.8 / Sonnet 5. (Anthropic)
Models & Tools
- OpenAI cleared for a broad GPT-5.6 launch; public availability Jul 9 (Jul 8) Trending Policy — The Commerce Department's Center for AI Standards and Innovation approved a wide rollout of the GPT-5.6 series after weeks of national-security testing, and OpenAI expects Sol (flagship), Terra (balanced), and Luna (fast/cheap) to become broadly available Jul 9. It's the first frontier model to complete the White House's new voluntary ~30-day pre-deployment review (from the June executive order) and ship, so the review is now operational, not theoretical. Sol's API tariff is $5 in / $30 out per M tokensA 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 ↗; if you've been latency- or capability-bound on a vetted-partner allowlist, general access opens this week. (CNBC | Axios | OpenAI)
APIs & Pricing
- OpenAI Assistants API: hard shutdown Aug 26 (deadline Aug 26) Pricing — Deprecated a year ago (notice Aug 26, 2025) and now feature-matched by the Responses API; after Aug 26 the legacy endpoints error out with no degraded mode or extension. The replacement is Responses (run + output) plus Conversations (stored back-and-forth), which also unlock MCPModel Context ProtocolAn open standard that lets AI apps connect to data sources, tools and workflows through one common interface.MCP ↗, deep research, and computer-use. Note there's no automated Threads→Conversations migration — plan the backfill. (OpenAI Deprecations)
- Anthropic: Fable 5 exits subscription inclusion today (Jul 8, update) Pricing — The temporary 50%-of-weekly-limits inclusion that ran through Jul 7 has now lapsed; continued Fable 5 access requires enabling usage credits at $10/M input, $50/M output. Orgs that don't enable credits lose access entirely. Anthropic frames the metered window as temporary while it builds capacity to restore standard-plan inclusion. (Anthropic)
Industry & Business
- xAI completes its SpaceXAI rebrand; SPCX joins the Nasdaq-100 (Jul 6–7) Business — Musk's AI lab (Grok, plus X) has fully folded into SpaceX under the SpaceXAI name, debuting a new logo and X handle — the capstone of SpaceX's February acquisition. Separately, following June's record ~$75B IPO and ~$2T valuation, SpaceX (SPCX) joined the Nasdaq-100 on Jul 7, triggering index-fund inflows at a ~1–1.3% weight. Consolidates Musk's space, AI, and social products under one ticker. (Yahoo Finance | Gizmodo)
- Norm raises $120M Series C at a $1.2B valuation (Jul 7) Business — Khosla Ventures led the round (with Blackstone, Bain Capital Ventures, Craft, Coatue, Vanguard and others) into the ~3-year-old legal-AI company, now a unicorn having raised $260M+ total. Norm runs an AI-native law firm ("Norm Law") where supervisory AI agents do the work under human-attorney oversight, serving enterprise clients representing $30T+ in AUM. A marker of how fast agenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗ automation is moving into regulated, high-liability domains. (TechCrunch | Bloomberg)
- Zoom to acquire Common Room for AI go-to-market intelligence (Jul 2) Customer AI — Zoom is buying the Seattle GTM-intelligence startup (terms undisclosed) to feed buyer signals into its Revenue Accelerator: Common Room unifies CRM/product/marketing data with real-world buying signals to tell reps which accounts are in-market before a call. Customers include Anthropic, Atlassian, Notion, Okta, and Snowflake; the deal is expected to close in the coming weeks. Another sign the SDR/RevOps stack is consolidating around AI-native buyer intelligence. (Zoom | GeekWire)
Benchmarks & Evals
- Tencent Hy3 debuts on the 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 leaderboard at 57.9% (leaderboard updated Jul 7) Research — Tencent's Hy3 was added to the SWE-bench Pro board (harder, actively-maintained repos with no public ground-truth leakage) at 57.9% — mid-pack, well behind the Claude Fable/Opus and GLM/Qwen leaders but a data point on Chinese labs steadily filling out the agentic-coding field. Useful if you benchmark candidate models for CI/agent workloads. (SWE-bench)