xAI is running an unusually tight release schedule this summer. Elon Musk has put public timelines on the next two Grok models — Grok 4.6 and Grok 4.7 — while the company has already shipped a new voice model, Grok Voice Think Fast 2.0, that becomes the default for Grok’s voice API on August 5, 2026.
Quick facts
- Musk says Grok 4.6 arrives around August 7, 2026, roughly two weeks after Grok 4.5’s July 16 launch.
- Grok 4.7 is expected a few weeks after that, described by Musk as a larger, 2.1-trillion-parameter model.
- Grok Voice Think Fast 2.0 is live now via the xAI API, and becomes the default “grok-voice-latest” model on August 5, 2026, priced at $0.08 per minute of audio.
- Grok 4.5, the current flagship, is a 1.5-trillion-parameter model priced at $2 per million input tokens and $6 per million output tokens, positioned specifically for coding and multi-step agentic tasks.
What Musk actually said
The Grok 4.6 timeline didn’t come from a press release — it came from Musk replying directly to a post on X. On July 24, 2026, Musk confirmed, “Grok 4.6 in 2 weeks and Grok 4.7 in 4 weeks.” He followed up on July 28 with more specifics, according to reporting from American Bazaar: Grok 4.6 is a 1.5-trillion-parameter model built around improved supervised fine-tuning and reinforcement learning, targeting an August 7 release, while Grok 4.7 will scale up further to 2.1 trillion parameters and improve on 4.6 across the board, with the tradeoff of being somewhat slower to serve despite better token efficiency.
Worth flagging: not every outlet agrees on the exact numbers. At least one other report describes Grok 4.6 as a 2-trillion-parameter model rather than 1.5 trillion, and xAI itself hasn’t published a full technical spec sheet for either model yet. Until xAI posts an official model card, treat any specific parameter count — including the ones above — as Musk’s stated intent rather than a confirmed final spec.
Where Grok 4.5 set the baseline
Grok 4.5 went public on July 16, 2026, built on what xAI calls a V9 foundation, and is explicitly positioned around coding and long, multi-step agentic work rather than general chat — reporting indicates it was trained in part on large volumes of real developer-agent session data through xAI’s integration with Cursor. According to figures cited by Basenor, it runs at 80 transactions per second and scores 29.0% on the SWE Marathon coding benchmark, ahead of the 26.0% reported for Claude Opus 4.8 — a comparison worth noting but treating as one outlet’s benchmark read rather than an independently verified head-to-head. Pricing sits at $2 per million input tokens and $6 per million output tokens.
Grok Voice Think Fast 2.0 is the part that’s actually shipping now
While the 4.6 and 4.7 timelines are still forward-looking, xAI’s own release notes confirm Grok Voice Think Fast 2.0 is available today, with meaningful gains in speech reasoning, transcription accuracy, and tool-use reliability over its predecessor. The company says the update is designed to improve performance across nearly all use cases without requiring any prompt changes. Starting August 5, 2026, the default “grok-voice-latest” routing moves from Think Fast 1.0 to 2.0 automatically; anyone who wants to stay on the older version needs to explicitly pin “grok-voice-think-fast-1.0” before that date. Pricing is transparent and flat at $0.08 per minute of audio.
xAI says early A/B testing on Starlink’s customer support line showed a meaningful increase in both sales conversion and support containment rates — a real-world enterprise use case rather than a benchmark score, though the specific figures haven’t been published.
Why xAI is moving this fast
A near-monthly cadence across model families is a deliberate competitive posture, not an accident. xAI is now backed by SpaceX, which announced its acquisition of xAI on April 17, 2026, giving the company deeper capital and infrastructure ties heading into a period where Google and OpenAI are both shipping updates on a similarly aggressive schedule. Fast, frequent releases let xAI respond to competitive pressure in smaller increments instead of waiting for a single, large flagship launch — at the cost of asking developers to keep re-benchmarking their own workloads every few weeks.
Common questions
Do I need to do anything before August 5? Only if you’re calling “grok-voice-latest” and specifically depend on Think Fast 1.0’s current behavior. Otherwise the upgrade to Think Fast 2.0 happens automatically with no code changes required.
Is Grok 4.6 available yet? No — as of publication it’s a stated target of around August 7, 2026, not a shipped model. Treat the date as Musk’s public commitment rather than a guaranteed release.
What’s actually different about Grok 4.5 versus a general chatbot? xAI has leaned specifically into coding and long, multi-step agentic workflows for this model line, rather than optimizing primarily for conversational chat — reflected in its Cursor integration and its benchmark focus on tasks like SWE Marathon.
How does the voice pricing compare? At $0.08 per minute, Grok Voice Think Fast 2.0 is priced flat per minute of audio rather than per token, which makes cost easier to predict for high-volume voice applications like customer support lines.
Key takeaway
If you’re building on Grok, the near-term action item is the voice model, not the frontier ones: the automatic switch to Grok Voice Think Fast 2.0 lands August 5, 2026, so pin the old version now if your application depends on Think Fast 1.0’s exact behavior. Grok 4.6 and 4.7 are worth watching but still subject to change — treat specific numbers as directional until xAI confirms them officially.
Leave a Reply