Gemini 3.5 Pro to be Released on July 17, xAI Grok 4.5 Starts Beta Testing
Google's Gemini 3.5 Pro will officially launch on July 17, featuring a significant leap in front-end and visual code generation capabilities, outperforming Anthropic's Fable 5 in several tests, though it still lags behind competitors in hardcore reasoning and complex engineering tasks.
Google DeepMind has abandoned the original 2.5 Pro base and opted for a completely new pre-training for Gemini 3.5 Pro, causing the release date to be pushed from the originally scheduled June 2026 to July 17.
Elon Musk announced that xAI's latest large model, Grok 4.5, has entered Beta testing internally at SpaceX and Tesla, with xAI planning to release a new large model trained from scratch every month for the remainder of the year.
Source: Public Information
ABAB AI Insight
Google DeepMind has been gradually catching up to OpenAI through iterations of the Gemini series. The new pre-training and release delay of 3.5 Pro reflect a cautious strategy adjustment in pursuit of cutting-edge performance, while xAI's rapid monthly iterations continue Musk's aggressive engineering culture, emphasizing speed and internal testing validation.
In terms of capital pathways, Google is enhancing multimodal capabilities through large-scale computational investments in retraining bases, while xAI relies on the ecosystems of SpaceX and Tesla for Beta testing and rapid iteration. Both are positioning themselves in the AI arms race with different rhythms: Google focuses on breakthrough points, while xAI emphasizes high-frequency validation and application landing.
This competition is reminiscent of the back-and-forth model iterations between OpenAI and Anthropic during 2023-2024, or the divergence between Google and Meta on open-source/closed-source routes. The large model industry is currently in a high-intensity iteration and ecological binding phase following the post-GPT-4 era.
Essentially, this is a combination of technological replacement and capital concentration: leading laboratories accelerate capability leaps through new pre-training and rapid iteration, while computational and data resources further concentrate among a few top players. Internal application scenarios (such as Tesla and SpaceX) provide a closed-loop feedback advantage for model training.
ABAB News · Cognitive Laws
The speed of model iteration determines the phase victory and defeat in the AI competition.
New pre-training, though costly, is a necessary path for performance leaps.
Real-world scenario testing is more effective in validating practical value than public benchmarks.