Grok 4.6 Delayed Past August 7 — Expected This Week
- What happened
- Grok 4.6 missed its August 7 launch date — no model card, benchmarks, or API listing published.
- Why it matters
- Musk's Aug 4 earnings call pointed to 'next week' (Aug 10–14) for release. Grok 4.7 (2.1T params) is weeks behind, and the aggressive three-model cadence is already slipping.
- What to do
- Use Grok 4.5 ($2/$6, available now) for production coding. Watch the xAI API console this week for Grok 4.6 availability.
Grok 4.6 has not shipped as of August 12 — Musk's "around August 7" target has passed with no model card, no API endpoint, no benchmark table, and no official statement from SpaceXAI explaining the delay. Our directory verdict stays pending until the model ships with independent evaluation.
What happened with Grok 4.6
Elon Musk first pegged Grok 4.6 to "around August 7" on July 27, then on the August 4 SpaceX earnings call pushed the window to "next week" — meaning August 10–14 (OrcaRouter(opens in new tab), 2026). That window is still open, but the silence is notable.
Musk posted about Grok Build V1.0, the coding harness, six times between August 5–8, ending with "Grok Build V1.0 is now released" — on the very day Grok 4.6 was supposed to land. Across those four days, neither Musk nor the company account mentioned Grok 4.6, 4.7, or Grok 5 once, per OrcaRouter's tracking.
What we know about Grok 4.6:
- Same 1.5T V9 foundation as Grok 4.5 with "significantly improved SFT & RL" — this is a post-training refinement, not a scale-up release
- Expected pricing: ~$2/$6 per 1M tokens, matching Grok 4.5 if serving costs hold at the same parameter count
- Grok 4.7 (2.1T params) follows "a few weeks" after 4.6, and has already drifted from "4 weeks" to "a few weeks" to "3–4 weeks" at each Musk statement
- LMSYS Arena plans to evaluate 4.6 the week after launch — the first genuinely third-party read anyone will get
Why it matters
SpaceXAI has been running an aggressive three-model-in-eight-weeks cadence targeting Anthropic's Opus line (StartupHub(opens in new tab), 2026). Grok 4.5 shipped July 8 and earned a conditional verdict from us — competitive pricing at $2/$6 and strong agentic coding performance on SWE-Bench Pro (64.7%) and Terminal-Bench 2.1 (83.3%), but mid-pack among frontier models on broader benchmarks.
Missing the first date in that cadence raises a real question: is this a routine slip, or a sign that the post-training improvements aren't delivering the promised capability jump? Without a model card or independent benchmarks, we have no way to know. Grok 4.6 stays pending — a pre-GA model that hasn't proven itself.
The parameter-count confusion that spread through early coverage doesn't help. Musk's July 24 remarks about a "2T model" led outlets to label Grok 4.6 as two-trillion-parameter. His July 27 correction was explicit: Grok 4.6 is the 1.5T model, and Grok 4.7 is the 2.1T model. This matters because it resets expectations — Grok 4.6 is a post-training step on the same foundation, not a bigger brain.
What changes for you
| If you are... | Your move |
|---|---|
| Building in production today | Use Grok 4.5 ($2/$6, ~80 TPS, available now) — it earned its verdict on shipped capability, not a roadmap slide |
| Evaluating 4.6 for a near-term switch | Watch the xAI API console this week (Aug 10–14); prepare a vendor-neutral endpoint so adopting 4.6 is a config change, not a re-integration |
| Waiting for the capability jump | Grok 4.7 at 2.1T is the release to watch — but with a timeline that has already drifted, don't plan sprints around it |
| Following the benchmarks | Arena's Agent leaderboard will publish live-session results within ~a week of 4.6's launch — weight that above any vendor slide |
FAQ
When will Grok 4.6 actually release?
Musk's most recent guidance points to the week of August 10–14. The August 7 target passed with no model card, but the company has gone quiet before a launch before. Check the xAI API console this week.
What's the difference between Grok 4.6 and Grok 4.7?
Grok 4.6 is a 1.5T-parameter post-training refinement — same foundation as Grok 4.5 with improved SFT & RL, targeting better agentic reliability at the same $2/$6 pricing. Grok 4.7 is the 2.1T scale-up, weeks behind, with higher capability and higher serving cost.
What to do
- 1 Use Grok 4.5 ($2/$6 per 1M tokens) for production coding and agentic tasks today — available now via the xAI API and Cursor
- 2 Check the xAI API console and developer release notes this week for Grok 4.6 availability
- 3 Prepare a vendor-neutral, OpenAI-compatible endpoint so adopting Grok 4.6 is a configuration change, not a re-integration
Affected tools & models
Never need to catch up again
The weekly delta — only verdict changes and act-now items. No digest filler.