July AI Battle Royale: Overseas GPT-5.6, Grok4.5, Gemini 3 Pro take turns appearing, and China's DeepSeek V4 official version is here too!

```

In July, the global competition for large AI models is entering a period of intensive releases, with leading players at home and abroad unveiling new products almost simultaneously.

OpenAI's flagship model GPT-5.6 Sol is scheduled for official release this Thursday. Elon Musk's SpaceX AI announced that Grok 4.5 will also be open to the public tomorrow. Google's Gemini 3.5 Pro is reportedly launching on July 17, while the official version of China's DeepSeek V4 also targets mid-July.

Within just a few weeks, several top-tier large models are being released in quick succession, directly intensifying competition in the AI industry and having far-reaching effects on the commercialization paths and pricing strategies of related companies.

For the market, this round of concentrated releases is not only a competition of technical capabilities, but also a full-scale battle in API pricing, inference efficiency, and ecosystem integration. The core issue for investors is: who can find the optimal balance between performance and cost, and be the first to turn technical advantages into business moats.

GPT-5.6 Sol: Flagship Positioning, Phased Access

OpenAI CEO Sam Altman announced on social media on Wednesday that GPT-5.6 Sol will be officially released this Thursday. This is the flagship product in the GPT-5.6 series, featuring a brand new ultra multi-agent mode and max inference power, and has set new best records in core benchmark tests such as coding, biology, and cybersecurity.

In terms of pricing, Sol adopts a fee of $5 per million tokens for input and $30 for output, ranking as the top tier among the three GPT-5.6 series products. OpenAI also announced that Sol will launch on Cerebras hardware in July, achieving inference speeds of up to 750 tokens per second.

Regarding release strategy, OpenAI is taking a phased approach: initially, only selected trusted partners will have access to the API and Codex, with plans to roll out the full GPT-5.6 series to a wider audience in the coming weeks. This strategy controls service pressure while also giving competitors some response time.

Grok 4.5: Musk Accelerates AI Strategy Integration

Elon Musk announced on social media that, following strong positive feedback from beta testing clients, SpaceX AI will open Grok 4.5 to the public tomorrow. Musk stated that the model is Opus-class, but faster, with higher token efficiency and lower cost.

According to previously disclosed information, Grok 4.5 is built on a V9 base model with 1.5 trillion parameters, and its supplementary training incorporated data from the AI programming tool Cursor. Early evaluations show that its performance is close to or may even surpass Anthropic's flagship model Claude Opus, with reinforcement learning continually optimizing the model.

Of note, SpaceX AI also plans to jointly launch the first collaboratively developed AI model with programming tool company Cursor, directly challenging Anthropic and OpenAI. According to an internal memo obtained by The Information, the release was previously delayed as both companies sought to further boost runtime efficiency, but the model is expected to rival Anthropic's Opus 4.8 and OpenAI's GPT-5.5 on certain benchmarks.

These developments come as SpaceX pushes forward with a $60 billion all-stock acquisition of Cursor, marking an acceleration in the integration of Musk's AI roadmap. Musk also revealed that SpaceX will launch new models trained entirely from scratch every month for the rest of this year.

Gemini 3.5 Pro: Rebuilding the Core, Betting on Quality First

On Google's side, leaked information indicates that Gemini 3.5 Pro will officially release on July 17. According to tech media Geeky Gadgets, Google DeepMind abandoned the original 2.5 Pro core, opting for all-new pretraining for Gemini 3.5 Pro, delaying the launch from the initially scheduled June 2026 to July 17.

In terms of capabilities, Gemini 3.5 Pro’s front-end and visual code generation have reportedly made a leap, outperforming Anthropic’s Fable 5 in multiple tests, but still lagging behind competitors in hard-core inference and complex engineering tasks. This decision is seen by outsiders as Google actively choosing quality over speed to address dual pressure from OpenAI’s GPT-5.6 and Anthropic’s Fable 5.

Additionally, Google is said to be developing a visual model called Nano Banana Pro based on this new foundation, targeting OpenAI’s GPT-Image 2, signaling Google’s intention to compete on both text-code and image-generation fronts simultaneously.

DeepSeek V4: Peak-Valley Pricing, Accelerated Inference in Parallel

On the domestic front, the DeepSeek team announced on June 29 that the official version of V4 is planned for release in mid-July, along with the introduction of a peak-valley pricing strategy. According to the published price list, API prices during peak hours will be twice the normal rate, while off-peak remains the same as the current V4 API pricing. Peak hours are defined as 9 am to 12 pm and 2 pm to 6 pm daily; the company says the move aims to allocate resources more rationally and improve service stability.

Technically, on June 27 DeepSeek, in collaboration with Peking University, released the inference acceleration framework DSpark and open-sourced the full-stack speculative decoding toolchain DeepSpec, with the paper authored by company founder Liang Wenfeng. Tests show that after DSpark deployment, V4-Flash single-user generation speed increased by 60%-85%, V4-Pro by 57%-78%, with the effect fully verified in online services. This is also DeepSeek’s first public open-source technology release since securing 50 billion yuan in financing.

For API users, peak-valley pricing will directly raise usage costs during working hours; for developers, the significant increase in inference speed may partially offset cost pressures in high-concurrency scenarios and further lower the threshold for inference optimization deployment.

Risk Warning and DisclaimerMarkets involve risk, and investment requires caution. This article does not constitute personal investment advice, nor does it take into account the specific investment objectives, financial situation, or needs of any particular user. Users should consider whether any opinions, views, or conclusions in this article fit their specific circumstances. Investments made accordingly are at your own risk. ```