Otimização de Custos de LLM

Reduza os custos de LLM em 50–90% com a combinação certa de cache, seleção de modelo, otimização de prompt e auto-hospedagem.

Quando Usar Esta Habilidade

Use esta habilidade quando:

• Os gastos com API de LLM estão crescendo mais rápido que a receita
• Você precisa atribuir custos de IA a equipes, produtos ou clientes
• Implementar cache para evitar chamadas redundantes de LLM
• Decidir quando mudar de provedores de API para modelos auto-hospedados
• Otimizar o comprimento do prompt sem sacrificar a qualidade

Alavancas de Custo por Impacto

Rastreie os Custos Primeiro

# Use LiteLLM's cost tracking (automatic per-model pricing)
import litellm

response = litellm.completion(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Hello"}],
)
cost = litellm.completion_cost(response)
print(f"Cost: ${cost:.6f}")

# Add custom cost callbacks
def log_cost(kwargs, completion_response, start_time, end_time):
    cost = kwargs.get("response_cost", 0)
    model = kwargs.get("model")
    user = kwargs.get("user")
    # Send to your analytics DB
    db.record_cost(user=user, model=model, cost=cost)

litellm.success_callback = [log_cost]

Dimensionamento Correto do Modelo

# Route by task complexity — don't use GPT-4o for everything
def get_model_for_task(task_type: str) -> str:
    routing = {
        "classification":     "gpt-4o-mini",      # ~30× cheaper than gpt-4o
        "summarization":      "gpt-4o-mini",
        "extraction":         "gpt-4o-mini",
        "simple_qa":          "gpt-4o-mini",
        "complex_reasoning":  "gpt-4o",
        "code_generation":    "claude-sonnet-4-6",
        "creative_writing":   "claude-opus-4-6",
    }
    return routing.get(task_type, "gpt-4o-mini")

# Cost comparison (per 1M tokens, 2025 approx.)
# gpt-4o-mini:          input $0.15 / output $0.60
# gpt-4o:               input $2.50 / output $10.00
# claude-sonnet-4-6:    input $3.00 / output $15.00
# llama-3.1-8b (self):  ~$0.05–0.10 all-in (GPU amortized)

Cache de Prompt (Lado do Provedor)

# Anthropic — cache long system prompts (saves 90% on cached tokens)
import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    system=[
        {
            "type": "text",
            "text": "You are a helpful assistant.",
        },
        {
            "type": "text",
            "text": open("large-context.txt").read(),  # large doc
            "cache_control": {"type": "ephemeral"},     # cache this!
        }
    ],
    messages=[{"role": "user", "content": "Summarize the key points."}],
)
# First call: full price. Subsequent calls: 90% discount on cached part.
print(f"Cache read tokens: {response.usage.cache_read_input_tokens}")

# OpenAI — prompt caching is automatic for repeated prefixes >1024 tokens
# No code change needed; check usage.prompt_tokens_details.cached_tokens

Agrupamento com a API em Lote da OpenAI (Desconto de 50%)

import json
from openai import OpenAI

client = OpenAI()

# Prepare batch requests
requests = [
    {
        "custom_id": f"task-{i}",
        "method": "POST",
        "url": "/v1/chat/completions",
        "body": {
            "model": "gpt-4o-mini",
            "messages": [{"role": "user", "content": f"Classify: {text}"}],
            "max_tokens": 50,
        }
    }
    for i, text in enumerate(texts)
]

# Write JSONL file
with open("batch.jsonl", "w") as f:
    for req in requests:
        f.write(json.dumps(req) + "\n")

# Upload and create batch
batch_file = client.files.create(file=open("batch.jsonl", "rb"), purpose="batch")
batch = client.batches.create(
    input_file_id=batch_file.id,
    endpoint="/v1/chat/completions",
    completion_window="24h",
)
print(f"Batch ID: {batch.id}")  # poll status with client.batches.retrieve(batch.id)

Cache Semântico

import hashlib
import json
import redis
import numpy as np
from sentence_transformers import SentenceTransformer

r = redis.Redis(host="localhost", port=6379)
embed_model = SentenceTransformer("BAAI/bge-small-en-v1.5")

SIMILARITY_THRESHOLD = 0.92

…