Skip to content

Supported LLM Providers

KTTC supports multiple LLM providers for flexibility and cost optimization.

OpenAI

Setup

export KTTC_OPENAI_API_KEY="sk-..."

Supported Models

Model Best For Cost Speed
gpt-4 Complex translations $$$ Medium
gpt-4-turbo General use $$ Fast
gpt-3.5-turbo Simple translations $ Very Fast

Usage

kttc check source.txt translation.txt \
    --provider openai \
    --source-lang en --target-lang es

Python API

from kttc.llm import OpenAIProvider

provider = OpenAIProvider(
    api_key="sk-...",
    model="gpt-4",
    temperature=0.3
)

Anthropic (Claude)

Setup

export KTTC_ANTHROPIC_API_KEY="sk-ant-..."

Supported Models

Model Best For Cost Speed
claude-3-5-sonnet-20241022 Most tasks $$ Fast
claude-3-opus-20240229 Highest quality $$$ Medium
claude-3-haiku-20240307 Speed & cost $ Very Fast

Usage

kttc check source.txt translation.txt \
    --provider anthropic \
    --source-lang en --target-lang es

Python API

from kttc.llm import AnthropicProvider

provider = AnthropicProvider(
    api_key="sk-ant-...",
    model="claude-3-5-sonnet-20241022",
    temperature=0.3
)

GigaChat (Russian Provider)

Setup

export KTTC_GIGACHAT_CLIENT_ID="your-client-id"
export KTTC_GIGACHAT_CLIENT_SECRET="your-client-secret"

Supported Models

Model Best For
GigaChat-Pro Professional translations
GigaChat General use

Usage

kttc check source.txt translation.txt \
    --provider gigachat \
    --source-lang en --target-lang ru

Python API

from kttc.llm import GigaChatProvider

provider = GigaChatProvider(
    client_id="your-client-id",
    client_secret="your-client-secret",
    model="GigaChat-Pro"
)

Note: GigaChat is particularly good for Russian language translations.

YandexGPT

Setup

export KTTC_YANDEXGPT_API_KEY="your-api-key"
export KTTC_YANDEXGPT_FOLDER_ID="your-folder-id"

Supported Models

Model Best For
yandexgpt General translations
yandexgpt-lite Fast & cheap

Usage

kttc check source.txt translation.txt \
    --provider yandexgpt \
    --source-lang en --target-lang ru

Python API

from kttc.llm import YandexGPTProvider

provider = YandexGPTProvider(
    api_key="your-api-key",
    folder_id="your-folder-id",
    model="yandexgpt"
)

Provider Comparison

Cost Comparison

Provider Model Cost per 1M tokens Quality Speed
OpenAI GPT-4 $$$ ⭐⭐⭐⭐⭐ ⭐⭐⭐
OpenAI GPT-4 Turbo $$ ⭐⭐⭐⭐ ⭐⭐⭐⭐
OpenAI GPT-3.5 Turbo $ ⭐⭐⭐ ⭐⭐⭐⭐⭐
Anthropic Claude 3.5 Sonnet $$ ⭐⭐⭐⭐⭐ ⭐⭐⭐⭐
Anthropic Claude 3 Opus $$$ ⭐⭐⭐⭐⭐ ⭐⭐⭐
Anthropic Claude 3 Haiku $ ⭐⭐⭐⭐ ⭐⭐⭐⭐⭐
GigaChat GigaChat-Pro $$ ⭐⭐⭐⭐ (Russian) ⭐⭐⭐

Language Support

Provider English Spanish French German Chinese Russian
OpenAI ⭐⭐⭐⭐⭐ ⭐⭐⭐⭐⭐ ⭐⭐⭐⭐⭐ ⭐⭐⭐⭐⭐ ⭐⭐⭐⭐ ⭐⭐⭐⭐
Anthropic ⭐⭐⭐⭐⭐ ⭐⭐⭐⭐⭐ ⭐⭐⭐⭐⭐ ⭐⭐⭐⭐⭐ ⭐⭐⭐⭐ ⭐⭐⭐⭐
GigaChat ⭐⭐⭐⭐ ⭐⭐⭐ ⭐⭐⭐ ⭐⭐⭐ ⭐⭐⭐ ⭐⭐⭐⭐⭐
YandexGPT ⭐⭐⭐⭐ ⭐⭐⭐ ⭐⭐⭐ ⭐⭐⭐ ⭐⭐⭐ ⭐⭐⭐⭐⭐

Smart Routing

KTTC can automatically select the optimal provider and model based on text complexity:

kttc check source.txt translation.txt \
    --smart-routing \
    --source-lang en --target-lang es

How it works:

  1. Simple texts (< 100 chars) → Cheap model (GPT-3.5, Claude Haiku)
  2. Medium texts (100-500 chars) → Mid-range model (GPT-4 Turbo)
  3. Complex texts (> 500 chars) → Best model (GPT-4, Claude Sonnet)

Cost savings: Up to 60% reduction in API costs.

Benchmarking Providers

Compare providers for your specific use case:

kttc benchmark \
    --source test.txt \
    --providers openai,anthropic,gigachat \
    --source-lang en --target-lang es

Output:

Provider Benchmark Results:
┌──────────┬──────────┬────────┬──────────┐
│ Provider │ MQM Score│ Time   │ Cost     │
├──────────┼──────────┼────────┼──────────┤
│ OpenAI   │ 96.5     │ 2.3s   │ $0.045   │
│ Anthropic│ 97.2     │ 1.8s   │ $0.038   │
│ GigaChat │ 94.8     │ 3.1s   │ $0.032   │
└──────────┴──────────┴────────┴──────────┘

Provider Selection Guide

Choose OpenAI if:

  • You need broad language support
  • You want reliable, consistent quality
  • You're already using OpenAI for other tasks

Choose Anthropic if:

  • You need highest quality for critical translations
  • You want faster response times
  • You need detailed reasoning

Choose GigaChat if:

  • You're translating to/from Russian
  • You need local Russian compliance
  • Cost is a priority for Russian content

Choose YandexGPT if:

  • You're in Russia or CIS countries
  • You need Russian language specialization
  • You want local data residency

API Rate Limits

Provider Requests/min Tokens/min
OpenAI 3,500 90,000
Anthropic 4,000 100,000
GigaChat 1,000 50,000
YandexGPT 500 30,000

Note: Limits vary by subscription tier. Check your provider's documentation.

Troubleshooting

Rate Limit Errors

If you hit rate limits:

# Add delay between requests
kttc batch --file data.csv --delay 1.0

# Use smart routing to reduce API calls
kttc check source.txt translation.txt --smart-routing

Authentication Errors

Check your API keys:

# Verify key is set
echo $KTTC_OPENAI_API_KEY

# Test with verbose mode
kttc check source.txt translation.txt --provider openai -vv

Model Not Available

Some models may not be available in all regions:

# Try a different model
kttc check source.txt translation.txt \
    --provider openai \
    --model gpt-4-turbo