OpenAI upgrades Codex and debuts GPT-6 Decisions API

At its DevDay 2026 developer conference in San Francisco, OpenAI announced major upgrades to its Codex coding assistant. Codex now features reusable cloud development environments to help teams share configurations, a voice-controlled command-line interface, and a code review view in the desktop application. Additionally, a new Codex Security Cloud service automatically scans GitHub repositories for vulnerabilities and drafts fixes. This security tier grants subscribers access to OpenAI's Daybreak Blue models, which are specialized for malware analysis and defensive cybersecurity tasks.

For software automation, OpenAI added a Computer Use feature to its Agents API, enabling autonomous software operation. This API also gains multi-agent capabilities, Tool Search, and Context Compaction. Through a partnership with Amazon, these features will integrate into Bedrock Managed Agents on AWS. OpenAI also introduced the Decisions API, a limited preview built on the low-cost GPT-6 Luna model. Designed for quick classification and routing tasks, this API operates ten times faster than GPT-6 Luna does through standard API channels.

To address speed demands, OpenAI launched an Ultrafast tier that boosts token generation in Codex up to eight times, reaching 300 tokens per second, and speeds up the API by six times. However, this performance comes at six times the standard rate. For the GPT-6 Astra model, this translates to $60 per million input tokens and $300 per million output tokens. By comparison, Astra's Fast mode offers 2.5 times the speed for double the price, while the older GPT-5.3-Codex-Spark achieved over 1,000 tokens per second on Cerebras hardware but with lower benchmark scores.

The Ultrafast tier is available to Enterprise customers and subscribers of a new Pro 500 plan, which costs $500 per month. Meanwhile, OpenAI is reducing limits on its existing $200 Pro plan, cutting ChatGPT Work and Codex usage from twenty times to ten times Plus usage, and halving GPT-6 Pro messages from 200 to 100 weekly. Finally, the company noted that its new GPT-6.1 Sol model costs $2 per million input tokens and $10 per million output tokens, with an Ultrafast version arriving soon.