On August 7, OpenAI officially released its next-generation flagship model GPT-5, and announced that it will be immediately available to free and paying users as the default model on the ChatGPT platform, and will be available to enterprises (Enterprise) and education (Edu) in the coming week.) payment plan is fully launched.

This year's most anticipated model-GPT-5 is launched!

OpenAI positions GPT-5 as the company's "first all-in-one" system: combining its o-series capabilities in reasoning and long-term "test-time compute" with the GPT series's rapid response and interaction advantages into the same model and product experience.

Company CEO Sam Altman described GPT-5 as "the best model ever" and "an important step towards General Artificial Intelligence (AGI)" at the press conference and public channels, and emphasized the practicality of this release. Improvements and significantly reduced error rates.

According to OpenAI, GPT-5 achieved industry-leading results in programming and general reasoning tests: On SWE-bench Verified, GPT-5 scored 74.9% when the "chain-of-thought/ thinking" mode was turned on. On Aider Polyglot, GPT-5 scored 74.9%.(Multilingual Programming Ability Assessment) is 88%, and the company claims that the "hallucination rate" of GPT-5 has dropped significantly to about 4.8%, far below the double-digit levels previously reported in the GPT-4o and o series.OpenAI also demonstrated multiple improvements in GPT-5 in medical/health advice, mathematical/scientific reasoning, and code generation scenarios, and emphasized that the new model can handle more complex tasks in longer contexts (millions of tokens).

The release of GPT-5 caused a sensation in the capital market

After the release of GPT-5, the U.S. stock market quickly fermented after hours and overnight.Investors generally regard it as an accelerating signal for OpenAI's commercialization path. The market's short-term demand expectations for AI tracks, especially computing power, cloud services and enterprise-level software, have been significantly revised upward. ETFs and leading stocks in relevant sectors during the overnight trading hours. Buying increased significantly after the announcement.

As an important strategic partner of OpenAI, Microsoft is believed by the market to be the first to seamlessly integrate GPT-5 in flagship products such as Microsoft 365 Copilot and GitHub Copilot. This expectation has directly pushed up investors 'optimism about Microsoft's enterprise-level subscription revenue and customer stickiness. judgment.In overnight after-hours trading of U.S. stocks, although Microsoft's share price did not jump sharply, capital flow data showed that its buying share continued to be higher than the day's average for the two hours after the news was released.

In addition, Amazon, the parent company of AWS, unexpectedly gained market attention. Although the GPT-5 core is still hosted on the Microsoft Azure cloud, Amazon announced before and after the press conference that AWS will support some OpenAI open models, making its ability to compete in the AI ecosystem. Boundary was reassessed, pushing Amazon's share price up about 3.7% during the positive stock trading session, making it one of the most prominent technology blue chips of the day.

Sentiment in the chip and computing power infrastructure sectors also boosted significantly in the overnight market.GPT-5 's "thinking" mode and longer-context and multimodal processing capabilities mean that the computing power consumption of a single inference call will be significantly higher than that of previous models, which directly benefits the demand curve of GPUs and dedicated AI accelerator cards.

On representative targets such as Nvidia and AMD, although overnight after-hours gains were limited, the implied volatility in the options market rose rapidly, indicating that short-term traders have begun to bet on the upside in subsequent trading days.In addition, supply chain manufacturers of data center construction and high-performance computing (HPC) equipment were also driven by stock index futures trading in Asia and Europe, with some semiconductor equipment companies rising more than 2% overnight in the European market.

However, the market is also divided.

Some institutions immediately issued a research report reminding that although GPT-5 performed well in many benchmark tests, compared with GPT-4 's cross-generation transition, the overall performance improvement is more towards "directional enhancement" than an all-round revolution.This means that there is still uncertainty from technological breakthroughs to revenue redemption, especially as enterprise customers need to weigh marginal operating costs and performance improvements when deploying on a large scale.Since GPT-5 may significantly lengthen response time and increase call costs in inference mode, how to find a balance between pricing and computing power consumption will directly affect the profit quality of OpenAI and its partners.

American AI model "big melee"

Since the beginning of this year, the battle for big models in the United States has been in full swing, and the release of GPT-5 has pushed this battle to a climax.

At the end of February 2025, OpenAI launched GPT-4.5 (codenamed "Orion") as an iterative upgrade to GPT-4o, emphasizing a more natural dialogue experience and visual capabilities, and providing support for multiple languages for the first time.Soon after, OpenAI adjusted the pace of release of the large model in early April and released lighter o3 and o4-mini versions to warm up for GPT-5.

At the same time, Google continues to advance its Gemini series of models in early 2025.The Gemini 2.0 Flash default model was released in January 2025, followed by a more bridging Gemini 2.0 Pro (with the "Deep Think" thinking mode) and 1 million tokens of ultra-long context window support.Later, the multimodal and efficient Gemini 2.5 Flash and 2.5 Pro were launched, and Gemini 2.5 Flash-Lite and the open research version of Gemma 3 were launched in June, further expanding its technical coverage and developer accessibility.

Anthropic did not miss the competition.Since its first edition was released in early 2023, its Claude series has continued to iteratively promote capabilities improvement.Claude 3 will be launched in March 2024 (including Haiku, Sonnet, and Opus versions), of which Opus has the strongest reasoning capabilities.Claude 3.7 (Sonnet) was launched in February 2025 with a mixed reasoning mode; Claude 4 (Opus and Sonnet versions) was launched in May, bringing "Extended thinking" and parallel tool use capabilities; Claude Opus 4.1 was released in early August following GPT-5.

Musk's Gro k model is also active on the track.Gro k 3 was announced in February 2025, claiming to use 10 times computing power for training and supporting "Think" and "Big Brain" modes, the former being used for reasoning and the latter being used for complex problems.xAI also released the faster but slightly less accurate Gro k 3 mini.

In July, OpenAI began releasing timeline signals related to GPT-5.Techradar reported that it announced in July that it would hold a live broadcast of "LIVE5TREAM" on August 7, and previously released small open source models such as gpt-oss-120b and 20b.

Finally, on August 7, OpenAI officially released GPT-5.This is its first "unified model", which combines the deep reasoning capabilities of the o-series models with the rapid response performance of the GPT series.CEO Sam Altman called it a key step towards AGI for the world.