Grok 4.6 Set for Aug 7, Grok 4.7 Weeks Later

SpaceXAI just set the most aggressive shipping pace in frontier AI. Grok 4.6 — a model NextBigFuture reports targets ~2T parameters that Musk claims will beat Kimi K3 — launches around August 7. Grok 4.7, a larger model, follows within weeks. Two frontier models in a single month is a cadence no other lab has matched.

The confirmation came in a July 24 post on X: "Grok 4.6 in 2 weeks and Grok 4.7 in 4 weeks" (Musk, 2026(opens in new tab)). If both dates hold, SpaceXAI will have shipped three frontier models — Grok 4.5 (July 8), Grok 4.6, and Grok 4.7 — inside 45 days.

What happened

Musk set hard dates. On July 24, he posted on X(opens in new tab): "Grok 4.6 in 2 weeks and Grok 4.7 in 4 weeks." That translates to approximately August 7 for Grok 4.6 and August 21 for Grok 4.7.

Grok 4.6 is a direct successor to Grok 4.5's architecture. NextBigFuture reports it targets ~2T parameters, and Musk confirmed it as the "2T model" that is "better than our 1.5T in every way" (NextBigFuture, 2026(opens in new tab)). Musk claims it will beat Kimi K3 — the Chinese model from Moonshot AI reported at ~2.8T parameters that has been climbing the Arena leaderboard (NextBigFuture, 2026(opens in new tab)). Initial pre-training was expected to finish around July 20–27, 2026.

Grok 4.7 is a larger model following weeks after 4.6. NextBigFuture speculates a 4-6T parameter scale, but no official figure has been announced. Details are scarce — Musk's tweet is the only official word on 4.7.

Grok 4.5 already proved the architecture on coding benchmarks, leading Opus 4.8 on DeepSWE and Terminal-Bench 2.1 at 4.2× the token efficiency with just ~16K output tokens (SpaceXAI, 2026(opens in new tab)). Two more models within 30 days would extend that lead in shipping velocity.

Why it matters

No other frontier lab is shipping at this pace. If both dates hold, SpaceXAI will have launched three models — 4.5, 4.6, and 4.7 — in roughly 45 days. OpenAI (GPT-5.5 in May, no launch dates for the next) and Anthropic (Opus 4.8 in June) look deliberate by comparison.

Musk is explicitly targeting Kimi K3, the Chinese model reported at ~2.8T parameters from Moonshot AI. The strategic message is blunt: a Western lab can match China's best at a fraction of the parameter count.

The pricing angle amplifies the story. Grok 4.5 already undercuts every Western frontier model at $2/$6 per million tokens (input/output). If Grok 4.6 delivers Kimi K3-class performance at that pricing, it would be the first Western model to match China's price-to-performance ratio at the frontier.

Caveats apply. All performance claims are Musk's own — no independent benchmarks yet. Grok models remain unavailable in the EU. And Musk's launch projections have slipped before. The "around" qualifier on both dates leaves room for drift.

What changes for you

• Evaluating Grok 4.5? Hold for 4.6 benchmarks. The architecture is the same lineage, but the new training run at larger scale could meaningfully close the gap to the absolute frontier.
• Optimizing for cost efficiency? Grok's $2/$6 pricing already leads the Western field. If 4.6 hits K3-class performance at those prices, a model change from more expensive alternatives could cut per-task inference costs substantially.
• In the EU? Nothing changes — Grok models remain unavailable in the region.

FAQ

Is Grok 4.6 a new architecture? No. It is a direct successor to Grok 4.5's architecture, rebuilt through a fresh training run at larger scale (~2T parameters per NextBigFuture). The performance jump comes from training improvements and increased parameter count, not architectural changes.

Will Grok 4.7 replace 4.6 immediately? Likely not. With only a two-week gap, both models will coexist — 4.6 for cost-sensitive workloads and 4.7 for maximum capability. SpaceXAI still offers Grok 4.0 alongside 4.5.

When will independent benchmarks land? Expect within days of the August 7 launch. The LMSYS Arena and Artificial Analysis will be the first reliable third-party scores.