OpenAI’s Codex racked up 90 million installs in a single week. That’s not a cumulative lifetime number. That’s seven days of downloads.

The surge, powered by the company’s recently launched GPT-5.5 model, pushed Codex past both its own prior weekly benchmarks and rival Anthropic’s Claude code tool.

What’s driving the numbers

The timing isn’t coincidental. GPT-5.5’s rollout brought meaningful upgrades to the Codex platform, including a 400K token context window for the tool itself and API support for contexts stretching up to 1 million tokens.

Token efficiency also got a serious upgrade. GPT-5.5 reportedly uses approximately 40% fewer tokens per task compared to its predecessor.

API pricing for the new model sits at $5 per million input tokens and $30 per million output tokens.

Performance benchmarks tell a similar story. On NVIDIA systems, Codex powered by GPT-5.5 is delivering what’s described as a 35x lower cost per million tokens. Over 10,000 NVIDIA employees are reportedly experiencing significant improvements in both token generation speed and overall coding efficiency.

The competitive landscape just tilted

The 90 million install figure matters less as an absolute number and more as a relative one. Codex didn’t just grow. It overtook Claude code, which had been building steady momentum as Anthropic’s answer to AI-assisted programming.

The jump to a 1 million token context window at the API level is particularly significant. Most real-world software projects contain far more code than earlier models could process in a single session. Developers were forced to chunk their code into smaller pieces and hope the AI could stitch together a coherent understanding. That constraint is loosening considerably.

What this means for the crypto and tech ecosystem

The 40% reduction in token usage per task also has direct economic implications for any team running AI-assisted development at scale. Crypto projects, particularly those funded by token treasuries with volatile valuations, are perpetually cost-conscious.

There’s a flip side worth watching. As AI coding tools become ubiquitous, the volume of code being produced across all software, including smart contracts, will increase dramatically. The same tools that help developers build faster can also help them introduce bugs faster, particularly in security-critical environments like DeFi where a single overlooked edge case can drain a protocol.