Write code
Functions, scripts, modules and entire pages.
> Write a Telegram bot for booking clients> open code --ru_
Programming models without a VPNClaude, GPT-5.3-Codex, DeepSeek, Qwen3 Coder, Kimi, and Codestral in one window
They write and explain code, find bugs, build page layouts, and help with SQL and Excel. Pay only for responses, pay via SBP.
- Claude Opus 5.5$0.036
- Claude Sonnet 5.5$0.018
- GPT-5.3-Codex$0.023
- Kimi K2.7 Code$0.006
ClaudeCodexDeepSeekQwen
> you: Regex for a Russian phone number
gpt-5.3-codex ▸ ^(\+7|8)\d{10}$
[ok] answer ready · charged $0.023 · $10 gets you ≈ 432 such answers
> ls ./code
Code models in the aggregator
The models most often chosen for programming:
Claude OpusLarge projects, reviews and tricky bugs.Learn more →Claude SonnetEveryday code — fast and half the price of Opus.Learn more →GPT-CodexOpenAI models for development: GPT-5.3-Codex and Codex Mini.Learn more →DeepSeekStrong code at a very low price.Learn more →Qwen3 CoderAlibaba's open model for code.Learn more →Kimi K2.7 CodeLayout, interfaces and turnkey development.Learn more →Codestral and DevstralMistral models for code completion and editing.Learn more →GLM 5.3Complex development and long tasks.Learn more →
> code --choose
Which code model should you choose
Go by the task: there's no need to pay for the most expensive model where a simpler one will do.
- Whole large projectHolds a large codebase in mind and reasons carefully.Claude Opus 5.5$0.036per regular response
- Everyday codeTweaks and fixes — fast and inexpensive.Claude Sonnet 5.5$0.018per regular response
- Step-by-step developmentOpenAI model built specifically for programming.GPT-5.3-Codex$0.023per regular response
- Layout and interfacesStrong at building interfaces.Kimi K2.7 Code$0.006per regular response
- Code completionQuick response — for hints as you work.Codestral$0.0019per regular response
- As cheap as possibleHundredths of a cent per answer.DeepSeek V4.1 Flash$0.00022per regular response
> code --help
What code models can do
Find errors
They explain why it isn't working and suggest a fix.
> Why does this script crash?SQL and Excel
Queries, formulas, and macros.
> Formula: sales total by managerLayout
HTML, CSS, and UI components.
> Build a product card layout> price code --beta 1.0
How much do code models cost
No subscription: pay only for responses. The price is from $0.00022 to $0.09 per typical response.
Beta 1.0 prices may change
Response size
≈ a 1000-token request and a 700-token answer — a typical conversation
| Model | Request, $ per 1M tokens | Response, $ per 1M tokens | Context | One response | $10 is enough for | ||||
|---|---|---|---|---|---|---|---|---|---|
| Claude Fable 5.1 | $20 | $100 | 1M | $0.036 | $0.09 | $0.36 | ≈ 277 | ≈ 111 | ≈ 27 |
| Claude Opus 5.5 | $8 | $40 | 1M | $0.014 | $0.036 | $0.14 | ≈ 694 | ≈ 277 | ≈ 69 |
| Claude Sonnet 5.5 | $4 | $20 | 1M | $0.0072 | $0.018 | $0.072 | ≈ 1,388 | ≈ 555 | ≈ 138 |
| GPT-5.3-Codex | $3.5 | $28 | 400K | $0.0095 | $0.023 | $0.084 | ≈ 1,058 | ≈ 432 | ≈ 119 |
| GPT-5.1-Codex-Mini | $0.5 | $4 | 400K | $0.0014 | $0.0033 | $0.012 | ≈ 7,407 | ≈ 3,030 | ≈ 833 |
| GPT-6.1 Sol | $4 | $20 | 1M | $0.0072 | $0.018 | $0.072 | ≈ 1,388 | ≈ 555 | ≈ 138 |
| Gemini 3.8 Flash | $1.65 | $8.25 | 1M | $0.003 | $0.0074 | $0.03 | ≈ 3,367 | ≈ 1,346 | ≈ 336 |
| DeepSeek V4 Pro 0813 | $1.32 | $3.96 | 1M | $0.0016 | $0.0041 | $0.018 | ≈ 6,313 | ≈ 2,443 | ≈ 541 |
| DeepSeek V4.1 Flash | $0.0566 | $0.24 | 1M | $0.000089 | $0.00022 | $0.00093 | ≈ 112,384 | ≈ 44,523 | ≈ 10,720 |
| Qwen3 Coder Next | $0.24 | $1.6 | 262K | $0.00055 | $0.0014 | $0.0051 | ≈ 18,115 | ≈ 7,352 | ≈ 1,953 |
| Kimi K2.7 Code | $1.3424 | $6.7 | 262K | $0.0024 | $0.006 | $0.024 | ≈ 4,144 | ≈ 1,657 | ≈ 414 |
| Grok Build 0.1 | $2 | $4 | 256K | $0.0018 | $0.0048 | $0.024 | ≈ 5,555 | ≈ 2,083 | ≈ 416 |
| GLM 5.3 | $0.158 | $2.8 | 1M | $0.00089 | $0.0021 | $0.0069 | ≈ 11,268 | ≈ 4,721 | ≈ 1,456 |
| Codestral 2508 | $0.6 | $1.8 | 256K | $0.00072 | $0.0019 | $0.0084 | ≈ 13,888 | ≈ 5,376 | ≈ 1,190 |
| Devstral 2 2512 | $0.8 | $4 | 262K | $0.0014 | $0.0036 | $0.014 | ≈ 6,944 | ≈ 2,777 | ≈ 694 |
| Muse Spark Open | $2.5 | $8.5 | 1M | $0.0033 | $0.0084 | $0.037 | ≈ 3,030 | ≈ 1,183 | ≈ 270 |
The estimate is approximate: the cost of a response depends on the length of your request and the model's answer. The actual generation cost is deducted from your balance.
> cat ./how-to-start.txt
How to use code models in Russia
- 01
Sign up for P2D
One account opens all the aggregator's models — no need to register with each one separately.
- 02
Top up your balance via SBP
Balance in dollars. No foreign card or VPN needed, no monthly fee.
- 03
Choose an AI model
Open the AI aggregator, choose a model, and ask your question in Russian. The cost of the result is deducted from your balance.
> man code
Frequently asked questions about programming models
Which model writes code best?
For complex projects, people most often choose Claude Opus and GPT-5.3-Codex; for everyday tasks, Claude Sonnet; to save money, DeepSeek. In the aggregator you can ask the same question to different models and compare.
Can a model be connected to a code editor?
Right now the P2D aggregator works in a chat window: you send a question or a code snippet and get a response. There are no API keys for connecting to editors yet.
Is a VPN needed?
No. The P2D aggregator opens from Russia without a VPN, and the service itself sends the requests to the models.
How to pay?
Top up your balance in dollars via SBP with a Russian card. No foreign card needed, and there's no subscription or monthly fee — you're charged only for what you generate.
Other models in the aggregator
> start --code
Try code models right now
Sign up in a minute, top up via SBP, pay only for the result.