🗾ITmedia AI+ (日本)•Stalecollected in 40m
Codex adds flexible rate limit reset for paid users
💡Gain better control over your API usage and avoid workflow interruptions with new flexible rate limit resets.
⚡ 30-Second TL;DR
What Changed
Paid users can now accumulate rate limit reset credits
Why It Matters
This change provides developers with more predictable access to coding assistance, reducing downtime during intensive development sessions.
What To Do Next
Check your OpenAI dashboard to see if the new rate limit management options are available for your Codex API tier.
Who should care:Developers & AI Engineers
Key Points
- •Paid users can now accumulate rate limit reset credits
- •Users gain control over when to trigger the reset
- •Designed to improve workflow efficiency for AI coding tasks
🧠 Deep Insight
Web-grounded analysis with 17 cited sources.
🔑 Enhanced Key Takeaways
- •These rate limit resets can be "banked" and are typically usable for 30 days after being granted, sometimes offered as referral rewards for Plus and Pro plans.
- •The introduction of flexible resets follows OpenAI's April 2, 2026, transition of Codex to token-based credit billing, which provides clearer visibility into usage costs by directly mapping input, cached input, and output tokens to credits, replacing previous per-message pricing.
- •The flexible reset mechanism is integrated into a broader "real-time access engine" developed by OpenAI to manage continuous product access, combining rate limits, usage tracking, and credit balances to sustain performance amidst rapid adoption of services like Codex and Sora.
- •Despite the intent for flexibility, the actual implementation of rate limit resets may not always apply universally across all quota types or account states, and the user interface might not explicitly clarify the full scope of a given reset.
📊 Competitor Analysis▸ Show
| Feature/Product | OpenAI Codex | GitHub Copilot | Amazon Q Developer (formerly CodeWhisperer) | Google Gemini Code Assist (CLI) | Anthropic Claude Code |
|---|---|---|---|---|---|
| Core Functionality | Autonomous coding agent, multi-step tasks, code generation, debugging, testing, review | AI code completion, suggestions, chat, agents, code referencing | Real-time AI coding companion, code generation, security scans, app transformation | AI code completion, generation, agent mode, pull request reviews | AI code generation, large context window, agentic capabilities |
| Pricing Model | Bundled with ChatGPT plans ($0-$200+/month), token-based credit billing on rolling 5-hour window; API key for per-token billing | Free, Student, Pro ($10/month), Pro+ (5x Pro limits), Max (highest limits); session & weekly limits based on tokens/model multiplier | Free tier (limited agentic requests), Pro tier ($19/user/month) with increased limits | Free tier (60 req/min, 1000 req/day for individuals), Standard (1500 req/day), Enterprise (2000 req/day); model requests aggregated | Pro ($20/month, ~44K tokens/5hr), Max 5x ($100/month), Max 20x ($200/month); token-based limits |
| Rate Limit Management | Rolling 5-hour and weekly token-based limits; option to purchase credits or use banked resets; continues active turn even if limit hit | Session and weekly (7-day) limits based on token consumption and model multiplier; displayed in IDE/CLI; upgrade for higher limits | Limited agentic requests for free tier; increased limits for Pro tier | Requests per user per minute/day; aggregated across models; model fallback may occur; some users report hitting limits quickly | Token-based limits per 5-hour window; higher tiers offer increased token capacity |
| Benchmarks (SWE-bench) | 85.5% autonomous task completion (GPT-5.5-Codex) | 54% autonomous task completion | Null | Null (some users report Pro model limits after 4-15 large prompts) | 80.9% autonomous task completion (Opus 4.6) |
🛠️ Technical Deep Dive
- Codex operates on a large-scale transformer neural network architecture, descended from GPT-3 and specifically fine-tuned for code understanding and generation.
- The modern Codex functions as a cloud-based autonomous coding agent, executing tasks within a sandboxed, virtual computer environment to ensure isolation and security.
- Its access control system is a "real-time access engine" that integrates rate limits, real-time usage tracking, and credit balances to enable continuous product access and manage demand.
- The Codex App Server utilizes a bidirectional protocol to decouple the core agent logic from various client interfaces, including the CLI, VS Code extension, web app, desktop app, and third-party IDEs, all through a unified API.
- The core orchestration is an "agent loop" that iteratively manages interactions between users, language models, and tools, handling inference calls, tool execution, and conversation state.
- Technical optimizations include stateless request handling for Zero Data Retention, strategic prompt caching for linear performance, automatic context window management via intelligent compaction, and robust multi-turn conversation handling.
- Codex leverages the GPT family of models, including coding-specialized variants (e.g., GPT-5-Codex series), offering context windows up to 1 million tokens in ChatGPT-authenticated sessions.
🔮 Future ImplicationsAI analysis grounded in cited sources
Increased adoption of AI agents for complex, long-running software development tasks.
Flexible rate limits reduce friction for power users, enabling more continuous and extensive use of autonomous coding agents for end-to-end tasks without abrupt interruptions.
Intensified competition among AI coding tool providers to offer more flexible and transparent usage-based pricing models.
OpenAI's move to flexible resets and token-based billing, along with competitor responses, indicates a market demand for more predictable and controllable costs, pushing others to innovate their pricing and limit management.
Greater focus on user experience and workflow integration for AI coding tools.
Addressing user frustration with rigid throttling directly improves workflow efficiency, suggesting that providers will prioritize features that enhance seamless integration and control over AI assistance within development environments.
⏳ Timeline
2021-08
OpenAI Codex (language model, based on GPT-3) launched, powering the original GitHub Copilot.
2023-03-23
OpenAI deprecated the original Codex model API, integrating code generation into its general-purpose GPT-3.5 and GPT-4 models.
2025-05
OpenAI revived the Codex name for a new autonomous, cloud-based coding agent, marking an 'agentic reboot'.
2026-02-02
OpenAI launched the Codex desktop app for macOS, followed by a Windows version on March 4, 2026.
2026-04-02
OpenAI transitioned Codex to token-based credit billing on a rolling 5-hour window, replacing the previous per-message pricing model.
2026-04-28
OpenAI publicly announced a rate limit reset for all paid Codex plans to celebrate a 'good week' and enable more building with GPT-5.5.
📎 Sources (17)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: ITmedia AI+ (日本) ↗