[AINews] Codex usage up >10x in 6 months to 7M users, +1M in the past ~day; did Codex overtake Claude Code??
Summary
OpenAI's Codex and ChatGPT Work usage surged, reaching 7 million active users by July 13, 2026, marking a 10x growth in six months and an increase of 1 million users in approximately 24 hours following the GPT 5.6 Sol launch and usage limit adjustments. Concurrently, Prime Intellect released "verifiers v1," a new environment stack for agentic reinforcement learning, enabling more efficient long-horizon rollouts and claiming a 100B reasoning model trained in under two days on 6 H200 nodes. The AI landscape also saw a shift towards "cost per task" benchmarks for coding agents, with "harness/orchestrator" design becoming a key differentiator. A privacy controversy emerged regarding xAI's Grok Build CLI, which allegedly uploaded entire repositories, prompting xAI to emphasize Zero Data Retention. Further developments included enhanced interoperability for open models with vLLM, new quantization methods, and a renewed focus on continual learning and agent-centric AI engineering.
Key takeaway
For AI Engineers building agentic systems, prioritize robust harness design and evaluate models based on "cost per task" rather than just token pricing to optimize real-world performance and efficiency. Given recent privacy concerns, ensure your agent tools implement Zero Data Retention policies by default, especially when handling sensitive code or proprietary data. Investigate frameworks like Prime Intellect's verifiers v1 for scalable agentic RL infrastructure to support long-horizon rollouts and maintain control over institutional knowledge.
Key insights
Rapid AI tool adoption and agentic system evolution underscore a shift towards outcome-driven, efficient, and secure AI deployments.
Principles
- Agentic systems require robust environment stacks for efficient long-horizon rollouts.
- Cost per task is superseding token pricing as a key metric for agent performance.
- Harness/orchestrator design is a critical differentiator for agent products.
Method
Prime Intellect's "verifiers v1" splits environments into taskset, harness, and runtime, storing rollout traces as message DAGs to achieve O(n) growth for long-horizon multimodal rollouts.
In practice
- Evaluate coding agents based on "cost per task" rather than just token price.
- Prioritize agent harness design for task-specialized outcomes.
Topics
- Agentic AI
- LLM Inference
- Data Privacy
- Coding Agents
- Model Quantization
- Reinforcement Learning
Best for: CTO, VP of Engineering/Data, AI Architect, AI Engineer, AI Scientist, Director of AI/ML
Related on AIssential
See Counsel's argued verdicts on the open AI decisions leaders are weighing →
Editorial summary, takeaway, and curation by AIssential. Original article published by Latent.Space - Www.latent.space.