Openrouter Performance Tuning

What it does

When your agent uses it

• Building real-time applications
• Scaling request volume

Example prompts

• “openrouter performance”
• “openrouter latency”
• “openrouter speed”
• “/openrouter-performance-tuning”

Requirements

• Python 3
• A credential in OPENROUTER_API_KEY
• Compatibility (from SKILL.md): Designed for Claude Code
• Pre-approved tools (allowed-tools): Read, Write, Edit, Grep, Bash(python3:*)

Workflow steps

6 steps, taken from the first numbered list in SKILL.md.

1. 1Establish a baseline: run benchmark_model() from Benchmark Latency against your candidate models (e.g. openai/gpt-4o-mini vs…
2. 2Check the results against the Model Speed Tiers table to confirm each candidate sits in the right tier for your latency budget (200-500ms…
3. 3Switch user-facing paths to stream_completion() per Streaming for Lower TTFT and verify ttft_ms drops (typically 2-10x).
4. 4Move batch workloads to parallel_completions() per Parallel Request Processing, capping concurrency with asyncio.Semaphore…
5. 5Apply Connection Optimization — one shared client with timeout=30.0 and max_retries=2 instead of a new client per request.
6. 6Work through the Performance Optimization Checklist (set max_tokens, shrink prompts, consider :nitro variants and provider routing), then…

What it can do on your machine

Read from SKILL.md and the folder at commit cfae287. It shows what the files ask for, not the result of running them.

• Tool permissions

  Pre-approves these tools, so the agent can use them without asking each time:

  • Read
  • Write
  • Edit
  • Grep
  • Bash(python3:*)

  From allowed-tools in the SKILL.md frontmatter.

• Runs code

  No scripts in the folder and no shell commands in SKILL.md (its code samples are python).

  From the folder's file list and the shell code blocks in SKILL.md.

• Network

  Hosts in commands or code, which the agent is likely to contact:

  • openrouter.ai

  From URLs in SKILL.md, links to its own repository left out.

• Credentials

  Names these keys or tokens, usually read from environment variables:

  • OPENROUTER_API_KEY

  From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

• Compatibility

  Designed for Claude Code

  From compatibility in the SKILL.md frontmatter.

Context cost

Openrouter Performance Tuning loads about 2.4k tokens when it runs, and up to ~6.2k if it reads all its reference files. Until then it costs about 69 tokens; SKILL.md has 582 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it

~69

When it runs · the whole SKILL.md, loaded when a task matches

~2.4k

With references · SKILL.md plus every file in references/, read only if the agent opens them

~6.2k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from jeremylongshore/tons-of-skills-marketplace at commit cfae287, republished under its MIT licence (© jeremylongshore). 582 words, ~2,429 tokens.

name

openrouter-performance-tuning

description

Optimize OpenRouter request latency and throughput. Use when building real-time applications, reducing TTFT, or scaling request volume. Triggers: 'openrouter performance', 'openrouter latency', 'openrouter speed', 'optimize openrouter throughput'.

allowed-tools

Read, Write, Edit, Grep, Bash(python3:*)

compatibility

Designed for Claude Code

version

1.20.0

license

MIT

author

Jeremy Longshore <jeremy@intentsolutions.io>

tags

saas, openrouter, performance, latency, optimization

import os, time, statistics
from openai import OpenAI

client = OpenAI(
    base_url="https://openrouter.ai/api/v1",
    api_key=os.environ["OPENROUTER_API_KEY"],
    default_headers={"HTTP-Referer": "https://my-app.com", "X-Title": "my-app"},
)

def benchmark_model(model: str, prompt: str = "Say hello", n: int = 5) -> dict:
    """Benchmark a model's latency over N requests."""
    latencies = []
    for _ in range(n):
        start = time.monotonic()
        response = client.chat.completions.create(
            model=model,
            messages=[{"role": "user", "content": prompt}],
            max_tokens=50,
        )
        latencies.append((time.monotonic() - start) * 1000)

    return {
        "model": model,
        "p50_ms": round(statistics.median(latencies)),
        "p95_ms": round(sorted(latencies)[int(len(latencies) * 0.95)]),
        "avg_ms": round(statistics.mean(latencies)),
        "min_ms": round(min(latencies)),
        "max_ms": round(max(latencies)),
    }

# Compare fast vs slow models
for model in ["openai/gpt-4o-mini", "anthropic/claude-3-haiku", "anthropic/claude-3.5-sonnet"]:
    result = benchmark_model(model)
    print(f"{result['model']}: p50={result['p50_ms']}ms p95={result['p95_ms']}ms")

def stream_completion(messages, model="openai/gpt-4o-mini", **kwargs):
    """Stream response for lower time-to-first-token."""
    start = time.monotonic()
    first_token_time = None
    full_content = []

    stream = client.chat.completions.create(
        model=model, messages=messages, stream=True,
        stream_options={"include_usage": True},  # Get token counts at end
        **kwargs,
    )

    for chunk in stream:
        if chunk.choices and chunk.choices[0].delta.content:
            if first_token_time is None:
                first_token_time = (time.monotonic() - start) * 1000
            full_content.append(chunk.choices[0].delta.content)

    total_time = (time.monotonic() - start) * 1000
    return {
        "content": "".join(full_content),
        "ttft_ms": round(first_token_time or 0),
        "total_ms": round(total_time),
    }

import asyncio
from openai import AsyncOpenAI

async def parallel_completions(prompts: list[str], model="openai/gpt-4o-mini",
                                max_concurrent=10, **kwargs):
    """Process multiple prompts concurrently."""
    semaphore = asyncio.Semaphore(max_concurrent)
    client = AsyncOpenAI(
        base_url="https://openrouter.ai/api/v1",
        api_key=os.environ["OPENROUTER_API_KEY"],
        default_headers={"HTTP-Referer": "https://my-app.com", "X-Title": "my-app"},
    )

    async def process(prompt):
        async with semaphore:
            response = await client.chat.completions.create(
                model=model,
                messages=[{"role": "user", "content": prompt}],
                **kwargs,
            )
            return response.choices[0].message.content

    return await asyncio.gather(*[process(p) for p in prompts])

# 10 requests in parallel instead of sequential
results = asyncio.run(parallel_completions(
    ["Summarize: " + text for text in documents],
    max_concurrent=5,
    max_tokens=200,
))

# Reuse client instance (connection pooling)
# BAD: creating new client per request
for prompt in prompts:
    c = OpenAI(base_url="https://openrouter.ai/api/v1", ...)  # New TCP connection each time
    c.chat.completions.create(...)

# GOOD: reuse single client
client = OpenAI(
    base_url="https://openrouter.ai/api/v1",
    api_key=os.environ["OPENROUTER_API_KEY"],
    timeout=30.0,           # Set appropriate timeout
    max_retries=2,          # Built-in retry with backoff
    default_headers={"HTTP-Referer": "https://my-app.com", "X-Title": "my-app"},
)
for prompt in prompts:
    client.chat.completions.create(...)  # Reuses HTTP connection

for model in ["openai/gpt-4o-mini", "anthropic/claude-3-haiku"]:
    r = benchmark_model(model, n=5)
    print(f"{r['model']}: p50={r['p50_ms']}ms p95={r['p95_ms']}ms avg={r['avg_ms']}ms")
# openai/gpt-4o-mini: p50=430ms p95=610ms avg=455ms
# anthropic/claude-3-haiku: p50=395ms p95=580ms avg=418ms

© jeremylongshore, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 10 other files (references) in skills/.curated/openrouter-performance-tuning of jeremylongshore/tons-of-skills-marketplace.

• SKILL.md
• references/batch-processing.md
• references/best-practices-summary.md
• references/caching-strategies.md
• references/configuration-tuning.md
• references/errors.md
• references/examples.md
• references/latency-optimization.md
• references/model-selection-for-speed.md
• references/monitoring-&-profiling.md
• references/request-optimization.md

Open the folder on GitHubat commit cfae287