Exchange Ranking
Top Exchanges
24h Volume Ranking
Popularity Ranking
Exchange BTC Balance
Proof of Reserves
Decentralized Exchanges
Funding Rate
Funding Heatmap
Liquidation Data
Max Pain
Long/Short Ratio
Whale L/S Ratio
Binance/Okex/Huobi L/S
Bitfinex Margin L/S
ETF Tracker
Solana ETF
XRP ETF
Hong Kong ETF
Bitcoin Treasuries
Crypto Reversal
Ethereum Reserves
HyperLiquid Wallet Analysis
Hyperliquid Whale Watch
Large Transactions
On-chain Movement
Bitcoin ROI
Stablecoin Market Cap
Options Analysis
News
Articles
Economic Calendar
Features
Wallet
Contract Calculator
Security
Collections
Watchlist
FollowingMusk expects to cool down on xAI's delayed release of Grok 4.7 model
Before the official release of Grok 4.7, Elon Musk has made modest adjustments to the model's expectations. He admitted on social media X that the model performed more favorably when compared to Anthropic's Opus 5.0, and was slightly less competitive than newer versions of Opus 5.1, OpenAI's Fable and even the Astra model. In addition, Musk also pointed out that multimodal performance still needs to be improved. These candid reviews are in sharp contrast to the popularity they usually create before a product launches.
How does Grok 4.7 perform?
In response to questions about how Grok 4.7 compares with competing products, Musk wrote: "It should be roughly the same as Opus 5.0, not Opus 5.1." The SpaceX CEO added that the model is better in some respects and slightly weaker in others.
Musk explained on Friday that Grok 4.7 will require "a few days of training" because the team may have imposed excessive penalties on response length during the reinforcement learning process. As a result, the new model "prematurely abandoned difficult tasks that could have been completed" and its self-review capabilities have not yet reached the required level.
Grok 4.7, originally scheduled for release on September 12, is reportedly a model with approximately 2.1 trillion parameters, supplemented by engineering data from SpaceX. If these reports are true, the latest model will be 40% larger than Grok 4.6, which has a parameter of approximately 1.5 trillion yuan.
Third-party evaluations showed that Anthropic's Claude Fable 5.1 scored 52.6% in the Terminal-Bench-Science test, and its public pricing is $10 per million input tokens and $50 per million output tokens. In contrast, Grok 4.7, which has not yet been released, has no confirmed benchmark scores or token pricing. As of now, claims that its costs may be 10 times lower than competing products have not been confirmed.
Musk has committed to AGI
It didn't take long for Musk to return to his ambitious release plan, outlining a rapid iteration sequence that follows the unreleased 4.7 model.
He said that Grok 4.8 is a model with 2.5 trillion parameters, trained based on xAI's new C++ software stack, and will complete training this week and begin intensive learning. He called this a "significant improvement" over version 4.7.
On this basis, Musk claimed that Grok 4.9 may reach the "Astra/Fable level," which refers to OpenAI's GPT-6 Astra and Anthropic's Fable series. He described the current 2.5 trillion-parameter model as superior to the earlier 2.1 trillion-parameter version trained with Jax, which had "errors that could only be corrected halfway through the run," and pointed to the upcoming 3 trillion-parameter version, which will be equipped with upgraded in-house training software and cleaner data.
The biggest claims are reserved for the top of the ladder. Musk said Grok 5 "may be better than any existing model" but added cautiously that "we will see and see." In response to another user's question about what model would provide specific capabilities, he only replied: "That would be Grok 5."
Musk did not provide a release date for the model and did not provide any evidence other than assertions. Currently, the measurable benchmark remains at Grok 4.6, while Opus 5.1, which Musk himself set as a benchmark, and OpenAI's Astra have entered the market, but Grok 4.7 is still waiting to be released.
