The 30-day clock we wrote about on June 16 has run out.

SpaceX's option to acquire Cursor for $60 billion was structured around the IPO date: exercise within approximately 30 days of the June 12 Nasdaq listing. SpaceX listed at $135, closed at $161, raised $75 billion — the largest IPO in history. The deal was expected to follow. Today is that day.

We asked three open questions on June 16. Here is where each one landed.

Open question 1: Would Grok V9-Medium publish benchmarks before the deal closed?

No.

Elon Musk confirmed training completed June 5. 1.5 trillion parameters, trained on real Cursor developer workflow data, public release targeting mid-June. That was five weeks ago. No public benchmark numbers on SWE-bench Verified, LiveCodeBench, or any independently reproducible standard evaluation have appeared in major coverage.

Grok Build 0.1 — a smaller dedicated coding model, separate from V9-Medium — has been in API beta since May 29. Signal of direction, not a substitute for V9-Medium evaluation numbers.

The benchmark gap is still open. The acquisition closes into it.

Open question 2: Would Fable 5 come back before the close?

Yes, and then some.

Claude Fable 5 and Mythos 5 were fully restored globally on July 1 after the US Commerce Department lifted export controls. The 19-day suspension that had complicated Cursor's model routing since June 12 is resolved.

And then some: Claude Sonnet 5 launched July 2 and became the default Free and Pro model. Terminal-Bench 2.1 score: 80.4%, which beats Opus 4.8 (74.6%) on the benchmark that most stresses sustained autonomous coding tasks. SWE-bench Verified: 85.2%. Introductory pricing: $2/$10 per million tokens through August 31 — 40% of what Opus 4.8 costs.

The Claude routing story inside Cursor is stronger today than it was on June 16. The model stack Cursor inherited for its new owner is not the same one that was there when the clock started.

Open question 3: What does the broader model landscape look like?

Materially different. GPT-5.6 Sol broadly launched July 8. Terminal-Bench 2.1: 91.9% — the highest on the chart, ahead of Sonnet 5's 80.4% and Fable 5's 83.4%. Chinese models now account for 30–46% of US enterprise API traffic on OpenRouter, up from 4.5% in H1 2025. GLM-5.2 hits 62.1% on SWE-bench Pro at $1.40 per million input tokens.

The competitive pressure on xAI is meaningfully higher than it was in June. The model V9-Medium needs to displace in Cursor's routing decisions is not a static target anymore.

What I think is happening with V9-Medium

Two scenarios.

Scenario A: xAI is timing a benchmark release to coincide with the acquisition announcement. "SpaceX acquires Cursor, here are V9-Medium's numbers" is a better story than releasing benchmarks quietly and then announcing the deal separately. Standard narrative management.

Scenario B: V9-Medium's coding performance doesn't clearly beat Sonnet 5 or Sol on standard evaluations. If the numbers sat obviously above both, there's no real reason to hold them — releasing strong benchmarks before the acquisition closes builds goodwill with the user base you're inheriting.

My prior leans toward Scenario B. Not because xAI is incompetent — the training data story, real Cursor sessions versus synthetic code, is legitimately interesting and worth watching — but because the silence is now five weeks long. Strong numbers get published. Awkward numbers get announced at controlled moments.

Maybe I'm wrong. I'd like to be wrong.

Source spread

• TechCrunch — SpaceX option to buy Cursor for $60B [builder] — Original deal structure; 30-day acquisition window timing.
• CNBC — SpaceX IPO [hype] — IPO close price, $75B raised, first-day performance.
• TechTimes — Grok V9-Medium training complete [builder] — Training confirmation; no benchmarks published alongside or since.
• Anthropic — Redeploying Fable 5 [builder] — Export control lift, July 1 global restoration timeline.
• CNBC — Chinese AI models' share of US enterprise API traffic [builder] — 30–46% OpenRouter traffic figure; competitive cost context.

Pros & cons

What's real:

• The deal closes with both parties financially healthy. SpaceX raised $75B on a successful IPO. There's no scenario this falls apart for financial reasons. If xAI takes the $10B work-for-hire path instead of acquisition, that's a strategic choice, not necessity.
• The model stack inside Cursor right now is genuinely strong. Fable 5 is back, Sonnet 5 is competitive on agent tasks, Sol is the best coding benchmark number we've seen. Whatever xAI does with routing, they're inheriting a high baseline.
• If V9-Medium publishes competitive numbers at deal close, the Scenario A reading is right and this article is premature about the concern. I'll update.

What deserves a side-eye:

• Five weeks, no benchmarks. That's not a typical hold-for-launch delay for a model training was confirmed complete on.
• The financial incentive to route Grok through Cursor remains: 67% of Fortune 500 users, Colossus infrastructure costs, the need to commercially validate xAI's model. Those facts haven't changed.
• The competitive bar V9-Medium needs to clear has risen since June. Sonnet 5 and Sol both shipped strong coding numbers in the time the acquisition clock was running.

What builders need to know

• The acquisition closes today, per the 30-day window from June 12. If you haven't evaluated your Cursor dependency, the window to do it without urgency has officially closed. Do it now anyway — knowing your options is still better than not knowing.
• Grok V9-Medium still has no public benchmarks. Run your own comparison: use Grok Build 0.1 (grok-build-0.1 in API beta) against your actual workloads alongside Sonnet 5 and Sol. Build the dataset before V9-Medium's launch pressures the routing defaults.
• Fable 5 is back and Sonnet 5 is available at introductory pricing. The comparison baseline you're evaluating against is stronger than it was in June. Sonnet 5 at $2/$10 per million and Fable 5 back in the stack is a good position going into whatever xAI announces.
• Watch V9-Medium's benchmark disclosure at launch. If it publishes against SWE-bench Verified and LiveCodeBench with numbers above Sonnet 5 and Sol, update your routing assumptions. If it avoids those benchmarks or uses novel internal metrics, that's the thing to be skeptical about.
• User-configurable model routing is officially intact per Cursor's public communications. Watch for quiet deprecation of non-Grok defaults over six to twelve months — not a sudden switch, but a drift in which model powers which context by default.

Further reading

• TechCrunch — SpaceX option to buy Cursor for $60B — original deal reporting, 30-day window
• CNBC — SpaceX IPO day — close price and $75B raised
• TechTimes — Grok V9-Medium distribution flywheel — training complete, June 5
• Anthropic — Redeploying Fable 5 — July 1 restoration details and conditions
• SWE-bench Verified — the canonical coding benchmark to demand from V9-Medium at launch