--- title: "Grok 4.7 Is Still Vaporware — and Musk Just Graded It Against a Model Anthropic Shipped Two Months Ago" date: 2026-09-17 tags: ["xai","grok","elon-musk","anthropic","opus-5","claude-code","industry"] categories: ["Industry"] summary: "Elon Musk promised Grok 4.7 in about ten days on September 2. Fifteen days past that, on September 14, he self-graded the still-unshipped model as 'roughly on par with Opus 5.0, not 5.1' — a model Anthropic actually shipped July 24. As of September 17, xAI's own docs still top out at Grok 4.6." --- ![Grok 4.7 Is Still Vaporware — and Musk Just Graded It Against a Model Anthropic Shipped Two Months Ago](/images/grok-4-7-vaporware-musk-opus-5-self-grade.png) On September 2, Elon Musk said Grok 4.7 was coming "in about 10 days." That put the target around September 12. It's now September 17, and [xAI's own model documentation](https://docs.x.ai/docs/models) still tops out at Grok 4.6 — no model card, no pricing, no benchmark table, no Grok 4.7 entry anywhere. In the gap between the promise and the ship date, Musk did something more interesting than staying quiet: he told the public how good the still-unreleased model is going to be, by comparing it to a competitor's model that's been generally available for nearly two months. ## The delay, in Musk's own words Around the original target date, Musk posted that Grok 4.7 needed "a few more days to cook," attributing the slip to a reinforcement-learning tuning problem — the model had been penalized too heavily for response length during training, which was making it give up on hard tasks too early and skip enough self-checking to hurt reliability. That's a real, specific technical explanation, and credit where due: it's more substantive than the usual one-line "soon" that precedes most AI model delays. Then, on September 14, replying to a user on X, [Musk offered a different kind of specificity](https://x.com/elonmusk/status/2099458047408013751): > "Grok 4.7 should be roughly on par with Opus 5.0, not 5.1. Better in some ways, worse in others. We need to fix multimodal performance. Grok 4.8 will be a noticeable improvement. Grok 4.9 is probably Astra/Fable class. Grok 5 maybe better than anything. We shall see." Read that sentence again. The model being compared favorably-ish to "Opus 5.0" is Grok **4.7** — not 4.8, not 4.9. Musk's own roadmap has 4.8 and 4.9 still ahead of it, each promising further jumps, before Grok 5 shows up as the model that might beat everything. Whatever Grok 4.7 actually turns out to be when it ships, its own creator has pre-announced that it isn't the exciting one. ## The model it's being compared to shipped two months ago [Claude Opus 5 launched July 24, 2026](/2026/07/claude-opus-5-launch-fable-class-coding-half-price/) — [Anthropic's own announcement](https://www.anthropic.com/news/claude-opus-5) put it at 97.0% on SWE-bench Verified, independently corroborated by vals.ai's public leaderboard, alongside a real pricing page, a system card, and immediate availability across the API, Claude Code, Bedrock, Vertex, and Foundry. Anthropic has since moved on: Opus 5.1 and the unrelated "One Claude" consumer consolidation both shipped in the two months since, while Claude Code's own changelog has racked up dozens of dated, itemized releases in the same window — [v2.1.274 shipped today](https://code.claude.com/docs/en/changelog), September 17, adding OTel spans and transcript self-healing for corrupted sessions. So the actual comparison on the table is: a model Anthropic shipped, benchmarked, priced, and has already iterated past twice, versus a model that exists only as a founder's tweet, being graded by that same founder against the older of the two Anthropic releases he could have picked. That's not a knock on Grok 4.7's eventual quality — it's entirely possible it lands well. It's a knock on the information environment xAI is asking developers to make decisions inside. "Roughly on par with Opus 5.0, not 5.1" is not a benchmark. It's vibes, delivered with the confident specificity of a benchmark, from the one person with the least incentive to lowball it. ## Two shipping cultures, one week apart This is the same contrast this blog keeps returning to, because the companies keep providing fresh evidence for it. Anthropic ships Claude Code changes as dated, versioned, itemized changelog entries — multiple times a week, often multiple times in a single day — whether the change is a new sandboxing feature or a one-line timeout fix. You can audit exactly what shipped, when, and verify it yourself against the public changelog. xAI's shipping cadence for Grok 4.7, by contrast, has been a sequence of X posts: a ten-day estimate, an explanation for why the estimate slipped, and now a self-administered grade against a competitor's two-month-old model, all without an actual model to point to. Developers evaluating either company's tools don't need to take a side based on branding. They can just ask which one gives them something to independently verify. As of today, only one of these two models has a docs page, a system card, and a leaderboard entry. The other has a founder's tweet promising it'll be fine, then admitting the genuinely good one is still two versions away. ## What to watch Grok 4.7 will presumably ship eventually — the RL fix Musk described is a plausible, fixable problem, not evidence of a deeper wall. When it does, the number that will matter is an independent SWE-bench Pro or Terminal-Bench run, not another Musk tweet. Until then, treat "roughly on par with Opus 5.0" as exactly what it is: a prediction about an unreleased model, made by the person selling it, about a model that's already one full version behind Anthropic's current shipping Opus. ## Sources - [xAI model documentation](https://docs.x.ai/docs/models) — checked directly September 17, 2026, no Grok 4.7 entry - [Elon Musk on X, September 14, 2026](https://x.com/elonmusk/status/2099458047408013751) — self-comparison of Grok 4.7 to Opus 5.0 - [Introducing Claude Opus 5](https://www.anthropic.com/news/claude-opus-5) — Anthropic, July 24, 2026 - [Claude Code changelog](https://code.claude.com/docs/en/changelog) — Anthropic, checked directly September 17, 2026 (v2.1.274)