Zukko Blog
Guides to Claude Code, Cursor, Codex and the Claude and GPT APIs: setup, token savings and troubleshooting.
Which model to use in Claude Code: slots, prices and the index
6 min · · IDE
Claude Code takes its models from several environment variables. Here's what to put in each slot by price and index, and what a session costs.
GPT-6 Sol or GPT-6.1 Sol: choosing by price, cache and index
6 min · · models
Two models from one line with the same context window and different prices. We compare the table and work out plain requests and a cached agent.
Data labeling with an LLM: a 50,000-row dataset on a cheap model
6 min · · автоматизация
How to label a dataset with a cheap model: instruction, JSON output, a Python script, sample review and what 50,000 rows cost.
Reasoning tokens: what model reasoning costs and how to keep it in check
6 min · · models
A short answer, yet thousands of output tokens in your stats — that's model reasoning. Where it comes from, what it costs, and how to control it in OpenAI and Anthropic format.
AI code review in GitHub Actions: a workflow with curl and gh CLI
5 min · · автоматизация
One YAML file, and every pull request gets a comment with the model's findings. The workflow step by step, token permissions, fork limitations, and the cost of a single review.
LLM API error handling: retries, 429, 503 and a fallback model
5 min · · errors
Not every error is worth retrying, and not every one can be survived on the same model. Zukko error codes, backoff between attempts, and a fallback model chain in code.
Anthropic API from Russia: moving your code and Claude Code to Zukko
6 min · · development
Anthropic SDK code and Claude Code move to the gateway with an address and key change. Where the /v1 trap is, which Claude models are available, and how caching works.
An OpenAI API alternative from Russia: what to change in your code
6 min · · development
Code built on the OpenAI SDK moves to an OpenAI-compatible gateway with very few edits. Which lines to change, what to check in streaming and the Responses API, and how to pay from Russia.
How to OCR a PDF with an AI model: pages to images, then one API call
6 min · · development
Scans without a text layer can't be parsed the usual way. Render pages to images, send them to a model, and work out what a document will cost.
Async OpenAI requests in Python: thousands of calls with retries and accounting
6 min · · development
A working script for a batch of thousands of requests: capped concurrency, retries only where they help, and token and cost accounting.
LLM model cascade: cheap model first, strong model on escalation
5 min · · development
How to build a cascade: the cheap model answers first, hard cases go to the strong one. Python code, threshold calibration and a cost formula.
What a Claude Code agent costs: a step-by-step session with cache
7 min · · агенты
What an agent's bill is made of, a 30-step session worked out on Opus 5.5 and Sonnet 5, where to see actual spend and how to set a ceiling.
Claude Fable 5.1 or Opus 5.5: when the pricier model is worth it
6 min · · сравнение
Fable 5.1 costs 8.7 times more per token than Opus 5.5, and its strength index is lower. When it is still worth testing, and how to tell whether it pays off.
Claude Opus 5.5 or GPT-6 Astra: prices, index and agent session cost
7 min · · сравнение
Two flagships compared on what can be checked: price list, cache, strength index and the cost of one long agent session, with formulas.
GPT-6.1 Sol or Claude Sonnet 5: price, cache, API format, choice
6 min · · сравнение
Sonnet 5 is cheaper on input and output, Sol on cache. We work out where each one costs less and show how to compare quality on your own tasks.
DeepSeek or Gemini Flash: V4.1 Flash vs 3.8 Flash on price and strength
6 min · · сравнение
Two popular low-cost models with a 1M context. What prices and the intelligence index say, where caching changes the math, and how to decide on your data.
The cheapest models for bulk tasks: prices, math and how to choose
6 min · · экономия
The six cheapest models in the catalog in one table: what tens of thousands of requests cost, and which criteria decide between them.
Model intelligence index: how to choose an AI model for a task
6 min · · models
Every model in the catalog has an intelligence index and a price. How to read the two together, and why the final word belongs to a test on your own tasks.
Qwen 3.8 Flash via API: setup, pricing and bulk text tasks
5 min · · models
The junior model of the Qwen 3.8 line for text tasks at volume. How to connect it, what it costs, and what to compare it with before you launch.
GPT-6 Luna via API: the low-cost GPT-6 for bulk text tasks
6 min · · models
The cheapest GPT-6 model in the catalog. How to connect it, what it can't take as input, and what labelling 100,000 messages costs.
Claude Fable 5.1 via API: setup, pricing and spend control
7 min · · models
Fable 5.1 is one of the most expensive models in the catalog. How to connect it, what a request costs, and how to cap spend with key settings.
DeepSeek V4 Pro via API: setup, pricing and image input
5 min · · models
How to connect DeepSeek V4 Pro, send it an image and estimate the cost. And why V4.1 Flash is the better deal for plain text.
DeepSeek V4.1 Flash via API: setup from Russia and pricing
5 min · · models
How to connect DeepSeek V4.1 Flash with a Zukko key and pay from Russia, what typical workloads cost, and which tasks the model suits.
Gemini 3.8 Flash via API: setup, pricing and bulk workloads
5 min · · models
How to connect Gemini 3.8 Flash with a Zukko key, what a million short requests cost, and which models to compare it with for bulk work.