If you are evaluating frontier model APIs for late-summer 2026, Grok 4.6's timeline deserves a dedicated watch: on July 28, Elon Musk replied to Vercel CEO Guillermo Rauch on X and said xAI would ship Grok 4.6 around August 7 with 1.5 trillion parameters, followed within weeks by a 2.1T Grok 4.7 — but xAI has not published benchmarks or pricing yet. This article maps the timeline, core specs, SFT/RL upgrade logic, competitive positioning, and controversy, and flags what still rests on a single social post rather than an official model card.
Why developers should care about the Grok 4.6 clock right now
The announcement looks like executive PR, but it hits engineering teams on three fronts:
- Selection windows are shrinking: Grok 4.5 shipped July 8; if Grok 4.6 lands August 7, a flagship's useful shelf life may be one month — stacked on the July Hugging Face breach and GPT-6 regulatory race, multi-model orchestration cost models need monthly recalculation.
- The source list is tiny: Grok 4.6/4.7 parameter counts and SFT/RL claims come from one Musk reply — no accompanying model card — unlike Grok 4.5's launch with 15 tracked benchmarks. Integrations must budget for date slip and spec changes.
- Open-weight shock is already here: Kimi K3's full open-weight release on July 26 (2.8T parameters) topped Frontend Code Arena at 1,679 points. If xAI compresses to three frontier models in roughly two months, competition shifts from raw scale to token efficiency and agent post-training quality.
Boundary first: Grok 4.6/4.7 dates and parameters below come from Musk's July 28 X post. xAI's blog and product pages have not confirmed them. "Musk time" historically slips by days to weeks — verify against official xAI channels before committing.
Timeline from Grok 4.5 through 4.6 and 4.7
| Date | Event |
|---|---|
| July 8, 2026 | xAI ships Grok 4.5 for coding and agentic work, co-trained with Cursor on real developer sessions; 500K-token context; $2/$6 per million input/output tokens |
| July 16, 2026 | Moonshot AI's Kimi K3 goes live as a hosted service |
| July 26, 2026 | Kimi K3 full open-weight release — 2.8T parameters, 1M-token context |
| July 28, 2026 | Musk posts Grok 4.6/4.7 roadmap replying to Rauch; same day 1,200+ employees at OpenAI, Anthropic, Google DeepMind, and Meta publish the "Pacing the Frontier" letter |
| ~August 7, 2026 (target) | Grok 4.6 planned — 1.5T parameters, SFT/RL upgrade focus |
| ~late Aug–early Sep 2026 (est.) | Grok 4.7 planned — 2.1T parameters; Musk says better than 4.6 except slightly slower to serve, with better token efficiency |
Core specs and competitive comparison
Grok 4.6 and Claude Fable 5.1 rows are pre-release claims — useful for timing and positioning, not head-to-head benchmarks.
| Model | Date | Parameters | Context | Pricing (in/out per 1M tokens) | Status |
|---|---|---|---|---|---|
| Grok 4.5 | Jul 8, 2026 | Undisclosed | 500K | $2 / $6 | Shipped |
| Grok 4.6 (announced) | ~Aug 7 | 1.5T | Undisclosed | Undisclosed | Tweet only, unshipped |
| Grok 4.7 (announced) | ~late Aug–early Sep | 2.1T | Undisclosed | Undisclosed | Tweet only, unshipped |
| Kimi K3 | Jul 26 open-weight | 2.8T (MoE) | 1M | $0.30 (cache hit) / $3 (miss) in, $15 out | Shipped |
| Claude Fable 5.1 (rumored) | Rumored Aug | Undisclosed | Undisclosed | Rumored Fable 5 ($10/$50) | Unconfirmed by Anthropic |
| GPT-5.6 Sol | Shipped | Undisclosed | Undisclosed | Undisclosed | Shipped |
Why xAI is emphasizing post-training, not just scale
Supervised fine-tuning (SFT) shapes behavior on curated examples; reinforcement learning (RL) teaches which multi-step action sequences work in agentic tasks. Musk's "significantly improved SFT & RL" framing continues Grok 4.5's playbook: co-training on real Cursor developer sessions yielded Terminal-Bench 2.1 at 83.3% and SWE-Bench Pro at 64.7%, using roughly 15,954 output tokens per SWE-Bench Pro task versus Opus 4.8's 67,020 — a 4.2x efficiency gap.
Grok 4.6's 1.5T jump is real scale-up, but Musk's Grok 4.7 framing — bigger at 2.1T, "better in every way except slightly slower to serve" — suggests two SKUs with different latency/quality trade-offs rather than one model for everything. The ~August 7 target lands almost 10 days after Kimi K3's open-weight shock; Musk commented "impressive" on Kimi K3 benchmark posts, and most analysts read the compressed xAI cadence as a direct response to intensifying competition from OpenAI, Anthropic, and Chinese labs.
Five-step pre-release checklist for developers
Before Grok 4.6 ships with a model card, manage integration risk so preview specs do not become production dependencies:
- Tag source tiers: Mark Musk's X post as directional; mark Grok 4.5's official model card as verified baseline — never mix them in SLA or cost models.
- Buffer the date: Add 7–14 days beyond August 7 for "Musk time" slip; keep Grok 4.5 or Kimi K3 as fallback on critical release paths.
- Recompute token efficiency, not just leaderboard rank: Grok 4.5 wins on fewer output tokens per agent task; if 4.6 continues that path, bills may beat "smarter but longer" rivals — wait for per-task token disclosures.
- Isolate eval environments: Preview periods often mean shared sandboxes or local benchmarks — state pollution and credential residue rise; sensitive agent chains need resettable hardware.
- Track August release clustering: Rumored Claude Fable 5.1 also targets August; when Grok 4.6/4.7 land in the same month, fix one prompt set and one hardware environment so cross-week comparisons stay valid.
What to flag before you trust this timeline
- Only one source so far: Musk's X reply — no xAI blog post, model card, or product page confirming specs or date.
- Benchmarks and pricing are blank: Unlike Grok 4.5's detailed launch card, Grok 4.6 has no independent evals or official documentation yet.
- xAI safety controversies on a separate track: In July 2026 xAI sued a user for allegedly using Grok to generate CSAM — the company's first such lawsuit, implicitly conceding safeguards can be bypassed. A January 2026 Common Sense Media report rated Grok among the worst chatbots for child-safety risks — relevant for enterprise vendor risk, not raw capability scores.
- Collides with an industry split on pacing: July 28 — same day as Musk's roadmap — OpenAI and Anthropic endorsed "Pacing the Frontier" at the corporate level; xAI is absent from the signatory list, underscoring divergent strategies on release speed.
Citable facts and sources
- Grok 4.6 target date: ~August 7, 2026 — Musk's July 28 reply to @rauchg, not an xAI official announcement.
- Grok 4.6 parameters: 1.5T — same single social post, unverified by third parties.
- Grok 4.5 pricing: $2/$6 per million input/output tokens; 500K context — xAI "Introducing Grok 4.5" and official model card.
- Grok 4.5 agent benchmarks: Terminal-Bench 2.1 83.3%, SWE-Bench Pro 64.7%; output token efficiency vs Opus 4.8 — xAI model card and third-party summaries (eesel.ai, felloai.com).
Parameters, pricing, and scores may change at launch. Verify against official pages. Sources:
xAI official blog — Introducing Grok 4.5
xAI Grok 4.5 official model card (media.x.ai)
Moonshot AI Kimi K3 open-weight repo (GitHub)
The Verge — Pacing the Frontier letter coverage
Frequently asked questions
When exactly is Grok 4.6 coming out?
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.
What's the difference between Grok 4.6 and Grok 4.7?
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 except for serving speed, trading some latency for better token efficiency.
Will Grok 4.6 beat Kimi K3 or Claude Fable 5.1?
Too early to tell. Grok 4.6 has no published benchmarks yet, Kimi K3 already has verified third-party scores, and Claude Fable 5.1 has not been officially confirmed by Anthropic.
How much will Grok 4.6 cost?
Unknown. Grok 4.5 launched at $2 per million input tokens and $6 per million output tokens — a reasonable reference, but xAI has not disclosed Grok 4.6 pricing.
Where will I be able to use Grok 4.6?
Based on Grok 4.5's rollout, expect Grok Build, the xAI API, and the xAI console first, with third-party integrations following — not confirmed for 4.6 yet.
Frontier model release density is climbing, and running multi-model agent evals on laptops or shared VMs means more state pollution, credential residue, and accidental lateral movement — reinstalling for every model swap is slow and hard to reproduce. If you need a resettable, dedicated Apple Silicon physical node with SSH and VNC access to isolate Grok vs open-weight comparisons, VMSPIN's day-rate cloud Mac mini is usually the safer bet; for longer-term cost math, see our buy vs rent break-even analysis. Whether Grok 4.6 lands on August 7 remains unknown — your lab environment does not have to bet on the calendar.