1.5T params · SFT/RL upgrade · Grok 4.7 roadmap · Kimi K3 pressure · Musk time elasticity
On July 28, 2026, Elon Musk replied to Vercel CEO Guillermo Rauch on X that xAI plans to ship Grok 4.6 around August 7 — a 1.5-trillion-parameter model with significantly improved SFT and RL — followed weeks later by Grok 4.7 at 2.1 trillion parameters. That is three frontier models in roughly two months after Grok 4.5, with no benchmarks, pricing, or model card published yet. If you are evaluating agent stacks or waiting on next-gen API pricing, this guide delivers a full timeline, Grok roadmap tables, competitor comparisons, SFT/RL context, flagged controversies, and a six-step pre-release runbook. Everything below is labeled by source tier. Data as of July 30, 2026.
Only one thing is on the record here: a tweet. Everything else — the 1.5 trillion parameters, the significantly improved SFT & RL, the follow-up Grok 4.7 at 2.1 trillion parameters — comes from that single X post, with no accompanying model card, benchmark suite, or pricing page the way xAI published for Grok 4.5. That distinction matters for anyone deciding whether to build around this release.
Here's the exact chain of events:
| Date | Event |
|---|---|
| July 8, 2026 | xAI ships Grok 4.5, its first model built specifically for coding and agentic work, co-trained with Cursor on real developer session data. It launched with a 500K-token context window, pricing at $2/$6 per million input/output tokens, and a published model card with 15 tracked benchmark scores. |
| July 16–26, 2026 | Moonshot AI's Kimi K3 goes from hosted preview to fully open-weight release (2.8T parameters, 1M-token context), immediately topping Hugging Face's trending chart and drawing praise from Musk himself in the comments on its benchmark posts. |
| July 28, 2026 | Musk posts the Grok 4.6/4.7 roadmap in reply to Rauch. Same day, more than 1,200 employees across OpenAI, Anthropic, Google DeepMind, and Meta publish the "Pacing the Frontier" letter, and both OpenAI and Anthropic formally endorse it at the corporate level. |
| ~August 7, 2026 (target) | Grok 4.6, 1.5T parameters, positioned as an SFT/RL upgrade rather than a raw scale-up. |
| Late August–early September 2026 (estimated) | Grok 4.7, 2.1T parameters, which Musk says will be "better than 4.6 in every way, except slightly slower to serve, albeit with even better token efficiency." |
Five common blind spots before you treat this timeline as a build plan:
The only source is a tweet: There is no xAI blog post, model card, or product page confirming Grok 4.6's specs or date — this is one executive's public statement, and xAI has no obligation to hit it.
"Musk time" has a track record: xAI, Tesla, and SpaceX timelines from Musk have historically slipped by days to weeks; treat "around August 7" as a target, not a guarantee.
Parameter count is not the whole story: Musk's choice of words — "significantly improved SFT & RL" — signals the upgrade is about post-training, not raw scale alone; Grok 4.7 will be bigger yet "slightly slower to serve."
Ignore Kimi K3 at your peril: Grok 4.6's target date lands almost exactly 10 days after Kimi K3's full open-weight release rattled the industry — the most plausible explanation for xAI compressing its release cadence.
Industry split on AI pacing: The same day Musk announced Grok 4.6/4.7, over 1,200 employees published "Pacing the Frontier" asking the US government to help slow automated AI development — xAI is notably absent from that list.
| Model | Date | Parameters | Focus | Status |
|---|---|---|---|---|
| Grok 4.3 Beta | Apr 17, 2026 | Undisclosed | Baseline | Shipped |
| Grok 4.5 | Jul 8, 2026 | Undisclosed (single SKU, not MoE) | Coding/agentic, co-trained with Cursor | Shipped, benchmarked |
| Grok 4.6 | ~Aug 7, 2026 | 1.5T | SFT/RL upgrade | Announced via tweet, unshipped |
| Grok 4.7 | ~late Aug–early Sep 2026 | 2.1T | Broad upgrade over 4.6, better token efficiency | Announced via tweet, unshipped |
All Grok 4.6/4.7 figures are unverified vendor claims from a single social media post — treat them as directional, not confirmed specs.
| Model | Vendor | Parameters | Context | Pricing (input/output per 1M tokens) | Source |
|---|---|---|---|---|---|
| Grok 4.5 | xAI | Undisclosed | 500K | $2 / $6 | xAI official |
| Grok 4.6 (announced) | xAI | 1.5T | Undisclosed | Undisclosed | Musk's X post (unverified) |
| Kimi K3 | Moonshot AI | 2.8T (MoE, ~16/896 experts active) | 1M | $0.30 (cache hit) / $3 (cache miss) input, $15 output | Moonshot + Hugging Face |
| Claude Fable 5.1 (rumored) | Anthropic | Undisclosed | Undisclosed | Rumored unchanged from Fable 5 ($10 in / $50 out) | 36kr, WinCentral leaks — unconfirmed by Anthropic |
| GPT-5.6 Sol | OpenAI | Undisclosed | Undisclosed | Undisclosed | OpenAI official |
Grok 4.6 and Claude Fable 5.1 rows are pre-release claims, not verified benchmarks — useful for gauging release timing and positioning, not for head-to-head performance comparisons. See our Grok 4.5 review and Kimi K3 open-weight explainer.
Musk's choice of words — "significantly improved SFT & RL" — signals xAI is doubling down on the same playbook that made Grok 4.5 competitive on agentic benchmarks: Grok 4.5 used roughly 15,954 output tokens per SWE-Bench Pro task versus Opus 4.8's 67,020, a 4.2x efficiency gap, largely credited to post-training on real Cursor developer sessions rather than sheer model size.
Whether you are an AI developer, a technical decision-maker, or an agent engineer, use these six steps to build an independent judgment framework before Grok 4.6 actually lands:
Verify source tier: Distinguish Musk's X post from an xAI official blog or model card — Grok 4.6 currently has only the former. Label every spec as announced/unverified until xAI publishes.
Build in Musk time buffer: Treat "around August 7" as a target window, not a hard deadline. Leave 1–2 weeks of slack in vendor selection and budget planning.
Anchor on Grok 4.5: 4.5 shipped with a full model card, 15 benchmark scores, and $2/$6 pricing — use that baseline to judge whether 4.6 is worth switching, not parameter count alone.
Cross-check Kimi K3 and Fable 5.1 rumors: K3 already tops Frontend Code Arena at 1,679 points and ranks third on Artificial Analysis's Intelligence Index; Fable 5.1 leaks point to an August window — three different positioning tracks, no head-to-head yet.
Understand what SFT/RL actually means: SFT shapes behavior on curated examples; RL teaches multi-step agent sequences via reward signals. Watch whether 4.6 preserves 4.5's token efficiency edge (15,954 vs 67,020 output tokens on SWE-Bench Pro).
Plan a hybrid routing strategy: Based on Grok 4.5 rollout, expect Grok Build, xAI API, and Cursor first — keep routine subtasks on proven models and reserve high-stakes architecture calls for verified frontier options until 4.6 pricing and benchmarks ship.
Supervised fine-tuning (SFT) trains a model on curated example outputs to shape its behavior; reinforcement learning (RL) uses reward signals to teach a model which action sequences actually work, which matters most for multi-step agentic tasks. Musk's choice of words — "significantly improved SFT & RL" — signals xAI is doubling down on the same playbook that made Grok 4.5 competitive on agentic benchmarks despite trailing rivals on raw intelligence scores: Grok 4.5 used roughly 15,954 output tokens per SWE-Bench Pro task versus Opus 4.8's 67,020, a 4.2x efficiency gap, largely credited to post-training on real Cursor developer sessions rather than sheer model size.
Grok 4.6's jump to 1.5T parameters is a real scale increase over Grok 4.5, but Musk's own framing of Grok 4.7 — bigger at 2.1T, "better in every way except slightly slower to serve" — suggests xAI is deliberately building two SKUs with different trade-offs rather than one model for everything. That's consistent with how other labs (Anthropic's Sonnet/Opus split, OpenAI's mini/full tiers) already segment for latency-sensitive versus quality-sensitive workloads.
Grok 4.6's target date lands almost exactly 10 days after Kimi K3's full open-weight release rattled the industry — Moonshot's model topped the Frontend Code Arena leaderboard at 1,679 points, becoming the first open-weight model to beat every closed model on that board, and ranked third overall on Artificial Analysis's Intelligence Index. Chinese financial outlets (Futu News, 21st Century Business Herald) reported that the release wiped an estimated $314 billion off combined OpenAI/Anthropic valuation expectations and $111 billion off Nvidia's market cap in a single day — striking figures, but they're analyst estimates relayed through Chinese media, not independently confirmed by any of the companies involved, so treat them as directional sentiment rather than hard fact. That backdrop is the most plausible explanation for why xAI is compressing its release cadence to three frontier models in roughly two months.
The only source is a tweet. There's no xAI blog post, model card, or product page confirming Grok 4.6's specs or date — this is one executive's public statement, and xAI has no obligation to hit it.
Benchmarks and pricing are blank. Unlike Grok 4.5's detailed model card at launch, Grok 4.6 has no independent evaluation or official model card yet — "1.5 trillion parameters" and "significantly improved SFT & RL" remain unverified vendor claims.
Background: xAI safety controversies. In July 2026, xAI sued a user for allegedly using Grok to generate child sexual abuse material (CSAM) — the company's first lawsuit of its kind, and one that implicitly concedes Grok can produce such content when safeguards are bypassed. A January 2026 Common Sense Media report had already rated Grok among the worst AI chatbots for child-safety risks. None of this affects Grok 4.6's technical capabilities, but it's relevant context for enterprises weighing vendor risk.
Background: industry split on AI pacing. The same day Musk announced Grok 4.6/4.7, over 1,200 employees at OpenAI, Anthropic, Google DeepMind, and Meta published "Pacing the Frontier," a letter asking the US government to help build tools to deliberately slow automated AI development — and both OpenAI and Anthropic endorsed it as companies. xAI is notably absent from that list, underscoring a real strategic divergence.
If Musk's timeline holds, Grok 4.6 and Grok 4.7 will land in the same month as a rumored Claude Fable 5.1 and just weeks after Kimi K3's open-weight shock. For teams evaluating models, that compresses the useful shelf life of any single flagship to a matter of weeks — which makes token efficiency and real-world task cost, not leaderboard rank alone, the more durable basis for a model choice.
Tracking multi-model API switching, agent orchestration pipelines, or Grok/Cursor joint dev workflows on a laptop often means unstable processes, no true 24/7 uptime, and compile-load bottlenecks; self-hosted VPS setups frequently lack Apple Silicon tooling and Metal compile chains. For production scenarios that need stable isolation, iOS CI/CD, and AI agent automation, VpsMesh Mac Mini M4 cloud rental is usually the better fit: unified memory suits large-context agent orchestration, and remote nodes can run 24/7 multi-model routing tests without polluting your local machine. See Mac Mini M4 rental pricing.
Musk said "around August 7, 2026" in an X post, but xAI has not officially confirmed a date. Treat it as a target that could shift. Follow xAI on X and x.ai for official announcements.
Grok 4.6 is a 1.5T-parameter model focused on SFT/RL post-training improvements. Grok 4.7, expected a few weeks later, is a larger 2.1T model that Musk says outperforms 4.6 across the board except for serving speed, where it trades some latency for better token efficiency. xAI appears to be building separate "faster" and "stronger" SKUs.
Too early to tell. Grok 4.6 has no published benchmarks yet, Kimi K3 already has verified third-party scores (including a leaderboard-topping result on Frontend Code Arena), and Claude Fable 5.1 hasn't even been officially confirmed by Anthropic. See our Kimi K3 open-weight explainer for competitor context.
Unknown. Grok 4.5 launched at $2 per million input tokens and $6 per million output tokens, which is a reasonable reference point, but xAI hasn't disclosed Grok 4.6 pricing. For agent test environments, see our help center.
Based on Grok 4.5's rollout, expect Grok Build, the xAI API, and the xAI console to get access first, with third-party platform integrations (like Grok 4.5's day-one availability in Cursor) following shortly after — but this isn't confirmed for 4.6 yet.