GPT-5.5 released three weeks ago, GPT-5.6 in internal testing; OpenAI and Anthropic begin subsidy battle, developers benefit

Just three weeks after GPT-5.5 launched, rumors about GPT-5.6 are already spreading.

Well-known leaker Leo revealed that GPT-5.6 development is now in full swing, with the first internal checkpoints being tested in recent days. An official unveiling is expected next month. Interestingly, traces of rollout mapping were found in OpenAI’s internal Codex logs — most calls still point to GPT-5.5, but one record clearly points to GPT-5.6. This suggests Codex may already be using it secretly for testing. Internal code names were also accidentally exposed: ember-alpha and beacon-alpha.

Netizens commented that this iteration speed is impossible to keep up with.

Speed pushed to the limit again — Codex to launch ultrafast mode

Even before GPT-5.6 arrives, OpenAI's product team can't wait any longer.

Leaker Chetaslua also revealed that this Thursday, OpenAI will introduce "ultrafast mode" on Codex, boosting response speed by 2 to 3 times, specifically for latency-sensitive tasks. At the same time, gpt-image-2, which leads the Image Arena rankings with +242 points, is undergoing A/B testing updates as well.

This is not the first time. When GPT-5.4 was released in March, the /fast mode already achieved a 1.5x speedup. GPT-5.3-Codex-Spark, powered by Cerebras chips, even increased reasoning speed to over 1000 tokens per second — 15 times faster than normal mode. This time, ultrafast mode directly accelerates the flagship main model — not a stripped-down version or a small model, but full-scale acceleration.

For developers, any area that involves "waiting" — such as Agent loops, long task pipelines, and browser automation — will feel entirely different.

By the way, last week the official ChatGPT account posted a sci-fi image, and the classic "Ask ChatGPT" on the interface was quietly replaced with "Message ChatGPT". Otto said "Call me maybe", which fueled speculation about ChatGPT integrating voice calls or launching hardware products.

Codex vs Claude Code: the subsidy war is officially underway

The most dramatic scene in Silicon Valley unfolded today.

Just before the launch of OpenAI's ultrafast mode, Anthropic made the first move — announcing that starting June 15, paid users will receive a 50% increase in monthly programming quotas, covering Claude Agent SDK, command-line tools, and Claude Code deeply integrated with GitHub workflows. They also released Opus4.7Fast mode, focused on faster advanced reasoning and a smoother long-context coding experience than Codex.

OpenAI did not keep people waiting. In response, they launched a powerful subsidy: any company that migrates to Codex from other platforms within the next 30 days will receive two months of free usage. Based on the $200/month Pro plan, that's equivalent to $400. Otto personally joined the campaign, stating that Codex is "the strongest AI programming product available on the market."

Just hours later, reports came in: 2,000 developers contacted OpenAI within 3 hours.

This face-to-face confrontation stunned the entire Silicon Valley. Developers, however, were happy — getting subsidies from two top companies like this is a rare opportunity.

The flywheel has already started spinning

Looking at these two events together, a trend more significant than any single piece of news is taking shape.

AI is accelerating its own evolution. GPT-5.3-Codex was OpenAI's first model that "participated in its own training." By GPT-5.5, 85% of OpenAI employees were using Codex to write code weekly. It's almost certain that GPT-5.6 was developed with deep involvement from GPT-5.5 — AI is helping OpenAI build stronger AI.

At the same time, the widespread adoption of programming tools is unleashing unprecedented engineering productivity. Codex already has 3 million weekly active users, and Claude Code's user base is also growing rapidly. Millions of developers now treat AI programming tools as a daily standard, and AI-generated code further feeds back into AI training and deployment. This cycle will only spin faster.

Otto said that OpenAI's goal is no longer just AGI, but directly aiming for ASI.

When model iteration speed is driven by AI itself, when AI programming tools become the infrastructure for AI development, and when two trillion-dollar companies use a "subsidy war" to accelerate this process —

The flywheel toward ASI has already begun to spin on its own.