Start in your Rate Cloud dashboard
Sign up or sign in, select a model, and get the steps to connect Codex.
The same interface
Keep working in the familiar Codex app instead of switching to a new AI client.
CODEX WITHOUT TOKEN METERThe same Codex.
The same Codex.
Different model.
No token limit.
Keep working in the familiar Codex, but send requests to GLM, DeepSeek, or Qwen through Rate Cloud. One command to connect, a fixed price for 30 days, and no cap on total tokens.
From 3 720 ₽ / 30 days · ∞ tokens · 4 concurrent slots
First 3 hours — free. After registering and confirming your email and Telegram address, you will be able to test all models with 1 parallel slot. Start free trial →
Codex● ● ●
Working on your project…
- ✓ Read 38 files
- ✓ Updated backend
- ✓ Running tests
- ● Checking implementation
Model: GLM 5.3Rate Cloud
Illustration · Tokens used this month: ∞Codex is not just GPT.
The model provider changes, not the application.
Codex supports custom model providers. Instead of the default OpenAI provider, you can route requests to a compatible Responses API and continue working through a different LLM.
Codex↓OpenAI → Rate Cloud↓GLM / DeepSeek / Qwen
Based on Codex setup documentation.
Codex remains Codex
We don't replace the application. We change the model that Codex sends requests to.
Your projects
Open repositories and working directories the same way you do now.
Terminal and tools
Codex continues to manage local tools and the client environment.
MCP
Connected MCP tools remain part of the Codex workflow.
Skills and instructions
The client's project instructions and skills continue to be used in the agent workflow.
Agentic work
Reading files, changing code, running commands, testing and multi-step work remain part of Codex.
Rate Cloud changes the LLM provider, not the Codex application itself.
Connect once. Then just work.
No need to manually figure out config.toml, API endpoints, and provider setup.
● ● ● Connecting Codex
1. Log into your account
Create an account or sign in at rate-cloud.com.
2. Choose a model
GLM 5.3, GLM 5.3 Flash, DeepSeek V4.1 Flash, DeepSeek V4 Pro, or Qwen 3.8 Max.
3. Connect Codex
Follow the instructions in the dashboard and use your Rate Cloud API key.
4. Work
New requests are routed through your chosen Rate Cloud model.
One thing changes
Your workflow stays in place.
Codex + OpenAI
Codex ↓ OpenAI ↓ GPT-6.1 Sol / Astra
Usage depends on your plan's allowance.
Codex + Rate Cloud
Codex ↓ Rate Cloud ↓ GLM / DeepSeek / Qwen
No limit on token count.
Codex is great. Limits run out before the work does.
Official plans include an allowance. It depends on the plan, model and type of work.
ChatGPT Plus
$20 / month. Codex included, but a usage allowance applies.
ChatGPT Pro
$100 / $200 / $500 per month. More included usage, but allowance remains.
Rate Cloud
3 720–4 640 ₽ / 30 days. No total token quota; up to 4 concurrent requests per subscription.
OpenAI rates and limits: Codex, Pro tiers, usage limits. Terms may change.
Your agent does not know that today is Sunday and the weekly limit is over.
It just needs to keep working.
Rate Cloud: 30 days of work without a token quota.Which model to replace GPT with?
There is no single best LLM for every task. Choose a model that fits your workflow.
GLM 5.3
4 640 ₽/ 30 days
A strong model for large coding tasks, long autonomous work, repository analysis, and managing a multi-step process.
Large repository · Planning · Tools
∞ tokens · 4 concurrent slots
Connect GLM 5.3 →GLM 5.3 Flash
4 070 ₽/ 30 days
A fast option for everyday development and a high volume of agent operations.
Speed · Many tasks · Subagents
∞ tokens · 4 concurrent slots
Connect GLM 5.3 Flash →DeepSeek V4.1 Flash
4 070 ₽/ 30 days
An option for heavy coding and repeated agent loops.
Code · Terminal · Tools
∞ tokens · 4 concurrent slots
Connect DeepSeek V4.1 Flash →DeepSeek V4 Pro
3 720 ₽/ 30 days
For complex reasoning and architecture analysis, where depth matters more than latency.
Architecture · Analysis · Complex tasks
∞ tokens · 4 concurrent slots
Connect DeepSeek V4 Pro →Qwen 3.8 Max
4 250 ₽/ 30 days
A model for large projects, long context, and sustained workloads.
Large context · Big projects
∞ tokens · 4 concurrent slots
Connect Qwen 3.8 Max →This is not "cheap models instead of GPT"
Public coding-agent tests show strong-tier results.
Terminal-Bench 2.1
Higher is better
The results were published by the model authors with different harnesses/configurations. A benchmark does not mean identical quality across all real-world tasks. Verify the methodology in Z.ai publications и DeepSeek materials. GPT-6.1 Sol is not shown here: no comparable result in this sample.
Not just one benchmark
Different tests measure different aspects of agent work.
GLM 5.3
Terminal-Bench 2.1 · 88.2
AutomationBench · 48.2
Agents’ Last Exam · 28.5
DeepSeek V4.1 Flash
Terminal-Bench 2.1 · 90.6
DeepSWE 1.1 · 74.2
AutomationBench · 54.8
Qwen 3.8 Max
Terminal-Bench 2.1 · 86.6
Toolathlon · 72.5
GDPval-AA v2 · 1739
GPT-5.6 Sol
Terminal-Bench 2.1 · 88.8
DeepSWE 1.1 · 72.7
AutomationBench · 45.8
Figures are from the published comparison table Z.ai; results from different tests cannot be added together or taken as a guarantee of outcome.
A strong coding model does not have to cost hundreds of dollars a month
Compare your specific scenario, not just the name of the plan.
Codex + Rate Cloud
GLM 5.3 · 4 640 ₽ / 30 days
DeepSeek V4.1 Flash · 4 070 ₽ / 30 days
∞ tokens · 4 concurrent requests · Codex as client
ChatGPT Pro
Pro 100 · $100 / mo
Pro 200 · $200 / mo
Pro 500 · $500 / mo
Codex included; allowance depends on plan
Fixed price for 30 daysIf you need intensive coding in Codex rather than the full ChatGPT Pro suite, Rate Cloud lets you pay only for LLM access for your coding workflow. These are different models and product bundles.
What if Codex really runs the whole month?
Take 500M tokens: 400M input + 100M output.
Rate Cloud / GLM 5.3
4 640 ₽ / 30 days
GPT-6.1 Sol API
~$1,800 at $2 input and $10 output per 1M tokens
GPT-6 Astra API
~$9,000 at $10 input and $50 output per 1M tokens
Fixed price for 30 daysComparison of the cost of processing the specified volume through the API, not of model equivalence. For OpenAI, the cost may differ with cached input, long context, and other modes. Check the current OpenAI API prices.
One Codex is good. Four processes — better.
One subscription lets you run up to four requests at once.
SLOT 1Lead agent
SLOT 2Code review
SLOT 3Tests
SLOT 4Research
2 subscriptions → 8 parallel slots
What you can leave Codex doing for a long time
While Codex runs, you do not have to watch the subscription's total token quota.
Feature development
Explore the project, implement features, run tests, and fix bugs.
Large refactor
Work with dozens of files without having to economize on the total number of tokens.
Bug hunting
Explore a large area of the project and test multiple hypotheses.
Testing
Generate, run and fix tests over several iterations.
Code review
Reserve a parallel slot to verify the result independently.
Subagents
Split research, implementation, and verification across processes.
Not a new AI editor. Not a new IDE. Not a new workflow.
If you're already used to Codex, you don't need to switch to another interface just to use a different model.
Your project↓Your Codex↓only changes hereRate Cloud↓GLM / DeepSeek / Qwen
Connect Codex
Choose a model. When the installer is ready, it will set up the rest.
RECOMMENDED FOR COMPLEX PROJECTS
GLM 5.3
4 640 ₽/ 30 days
∞ tokens · 4 concurrent requests · Codex compatible · API access
Buy and connect →View the instructions firstAfter purchase: a clear connection path
All the data you need and the instructions are available in your dashboard.
1. Your API key
Create and copy the key in your dashboard. Do not publish it.
2. Connection
Follow the Codex instructions in the Rate Cloud dashboard.
3. Operation
Open or restart Codex and work with your chosen model.
Not satisfied? Go back to OpenAI.
Rate Cloud is added as a separate provider: you can select the original OpenAI provider again in Codex settings.
Connection under your control
Setting up the provider does not require replacing the Codex application itself.
Separate provider
Rate Cloud is connected separately from your original OpenAI provider.
Your settings
Your MCP servers, skills and project instructions stay in Codex.
Your API key
Use the key from your dashboard only to authorize Rate Cloud requests.
Frequently asked questions
Honest about the model, client and limitations.
Can I try it for free?
Yes. After registration and confirmation of your email and Telegram address, all Rate Cloud models are free for the first 3 hours with 1 parallel slot. You can test them in Codex before buying a subscription.
Is this real Codex?
Yes. You keep using the Codex app. Rate Cloud changes the LLM provider instead of replacing the client.
Is this still GPT?
No. After connecting Rate Cloud, Codex uses your chosen GLM, DeepSeek, or Qwen instead of the OpenAI model.
Will the Codex interface change?
No separate editor is installed. The Codex interface stays familiar.
Will my skills remain?
Your skills, MCP and project configuration stay in place.
Will chat history remain?
Connecting Rate Cloud does not delete existing chat history.
Can I go back to GPT?
Yes. You can select the OpenAI provider again in the Codex settings.
Is there a token limit?
Rate Cloud does not set a total quota on input and output tokens during the 30-day subscription.
What is limited?
Up to four requests run simultaneously per subscription.
Can I work 24/7?
Rate Cloud has no 5-hour or weekly token quota. Actual volume depends on model speed and request duration.
What if I need 8 agents?
Two subscriptions of the same model give you up to eight parallel slots.
Which model is closest to top-tier GPT for coding?
For complex autonomous development start with GLM 5.3; for heavy coding and speed try DeepSeek V4.1 Flash. Quality depends on the task.
Keep Codex.
Remove the token counter.
Choose a model now. The connection command appears once the verified installer is released.
Connect Codex →From 3 720 ₽ / 30 days · ∞ tokens · 4 concurrent slots