OpenAI's 'GPT-6' Codenamed Astra Reportedly Launching in August! 10 Trillion Parameters, 'Doug' Model to Challenge Again by Year-End

A series of rumors about OpenAI's next-generation model have recently circulated on social platform X. Although the company recently suspended the launch plan for unknown reasons, leaks suggest that the new model, codenamed 'Astra'—widely speculated to be 'GPT-6'—could be released within this month.

As of now, OpenAI has not officially confirmed any details. All information comes from social media accounts and industry observers' speculation and secondhand reports.

The account 'ChrisGPT' claims that Astra's parameter count could reach approximately ten trillion. If true, this would be more than five times the estimated 1.8 trillion parameters of GPT-4.

The rumors also reference a post recently shared by OpenAI President Greg Brockman on X, which mentioned that GPT-4's training was completed on August 8, four years ago. Some users interpret this timing coincidence as a signal that GPT-6 might launch around the same period.

If the parameter scale indeed increases dramatically as rumored, the industry generally believes the new model should see significant improvements in logical reasoning and generalization capabilities. However, no independent credible sources have verified these figures, and everything remains speculative.

In addition to Astra, social media reports also point to another project codenamed 'Doug.' It is said to be far larger than the Fable 5 released earlier this year and is not GPT-6 itself, but rather an independent training outcome expected to launch by November.

Research firm SemiAnalysis previously mentioned this codename. Market analyst Pankaj Kumar speculates that 'Doug' will likely be trained using NVIDIA's (NVDA-US) next-generation Vera Rubin chips, which have not yet entered mass production.

If these rumors are true, OpenAI plans to launch two heavyweight models in the second half: Astra to dominate market attention in August, followed by 'Doug' to push technical benchmarks by year-end, aiming to solidify its industry leadership.

However, these details remain unconfirmed industry rumors.

Some commentators highlight a more significant industry phenomenon behind these rumors: since the release of GPT-4o in May 2024, OpenAI has not completed full-scale pre-training for a new cutting-edge model for over two years.

It is widely believed that updates such as o1, o3, and the GPT-5 series have mostly been post-training and reinforcement learning based on the existing GPT-4o foundation, rather than building a new model from scratch.

There are also reports that some internal projects were downgraded due to underperformance, but the company never abandoned its direction toward large-scale pre-training. It was only after Google (GOOGL-US) launched Gemini 3 last year that OpenAI accelerated its generational shift, leading to the verification project codenamed 'Garlic' and the larger-scale 'Doug' project.

If true, this aligns with the long-debated industry question: 'Has the scaling law failed?'

However, SemiAnalysis argues that as long as computing resources and data volume continue to grow, model performance can still improve as expected.

Amid these rumors, Google has also experienced major personnel upheaval. On August 5 local time, Jeff Dean, its chief scientist with 27 years of service, reportedly left to start a new venture with three core researchers.

On the same day, Demis Hassabis, co-founder of Google DeepMind and Nobel laureate, announced stepping down as CEO to become chairman. Alphabet's stock price dropped over 4% that day due to the news.

This personnel change is seen by some observers as a symbolic event, indicating that the previously 'three-way race' in generative AI has effectively narrowed to a direct confrontation between OpenAI and Anthropic.

In response to the potential threat from Astra, market rumors suggest Anthropic is planning to launch 'Fable 5.1' in August, priced the same as the current Fable 5, seen as a strategic countermove.

Some analysts believe Anthropic intentionally held back part of its technical capabilities when launching Opus 5, possibly to save strength for Fable 5.1, enabling a rapid response once OpenAI's new model launches.

However, this 'tactical product positioning' theory remains market speculation, and Anthropic has not officially confirmed any release plans or timelines.