ModelBeat is a multi-provider AI gateway that gives your team a single API key, unified billing, automatic failover, and cost-aware routing across 250+ models — without the hype about quality oracles we can't yet deliver.
Change your base URL and API key. That's it. Your existing OpenAI or Anthropic code works unchanged — ModelBeat handles the rest.
from openai import OpenAI
client = OpenAI(
api_key="mb-...", # one ModelBeat key
base_url="https://api.modelbeat.ai/v1", # ← one line changed
)
resp = client.chat.completions.create(
model="auto", # ModelBeat routes it
messages=[{"role": "user", "content": "Summarize this contract."}],
)
# See exactly which model handled it + cost
print(resp.model) # → "anthropic/claude-3-haiku"
print(resp.usage) # → tokens + cost breakdownimport OpenAI from "openai";
const client = new OpenAI({
apiKey: "mb-...", // one ModelBeat key
baseURL: "https://api.modelbeat.ai/v1", // ← one line changed
});
const resp = await client.chat.completions.create({
model: "auto", // ModelBeat routes it
messages: [{ role: "user", content: "Summarize this contract." }],
});
// Full transparency on routing decision
console.log(resp.model); // → "openai/gpt-4o-mini"
console.log(resp.usage); // → tokens + cost breakdowncurl https://api.modelbeat.ai/v1/chat/completions \
-H "Authorization: Bearer mb-..." \
-H "Content-Type: application/json" \
-d '{
"model": "auto",
"messages": [{"role": "user", "content": "Summarize this contract."}]
}'
# Response includes routing metadata:
# x-modelbeat-model: anthropic/claude-3-haiku
# x-modelbeat-cost: $0.000042
# x-modelbeat-routing: cost-tier-matchextra_body={"modelbeat": {"max_cost": 0.01, "task": "code"}} to override routing for specific requests. Ignored by anyone who doesn't need it.
Most routing products oversell quality prediction. We ship what's proven, instrument for what's next, and earn into the research bet with real data.
Rules-based cost-tier routing. Code fences → frontier model. Simple Q&A → cheap model. Image input → vision model. Delivers real cost savings on the 60% of traffic that's trivially classifiable.
Embed incoming prompts, compare to clusters of past prompts with logged outcomes. Route to whatever historically won that cluster. Gets smarter with your traffic — not someone else's.
A model trained on (query, candidate model, measured output quality) to predict the best model per input. This is what Martian and Not Diamond are building. We'll earn into it with data — not promise it on day one.
One key per project maps to a credit balance. Top up via Stripe. Hard budget caps and alerts so a runaway loop can't drain your balance silently. Transparent margin — no hidden 20–50% markups.
mb-sk-proj_abc123...
→ routes to 250+ models
Per-request logs: tokens, cost by type, model chosen, routing decision, latency. Queryable and exportable. Enterprises will ask — we're ready.
Provider goes down? Traffic reroutes instantly. Single providers rarely exceed 99.7% uptime — we give you 99.95%+ across the fleet.
Text, image, audio, video — ModelBeat inspects the request and routes to a capable model in the right modality automatically. No config required.
Prompt-injection screening, PII detection, content safety — on by default. Configurable per project: custom blocklists, vendor integrations, output schema validation.
Every response tells you which model handled it and why. "X% of traffic routed to cheaper models, $Y saved vs all-frontier." The metric that proves the cost story.
We're not trying to out-research Martian or out-breadth OpenRouter. We're the honest middle: production-grade gateway with real cost savings and no quality claims we can't back up.
| Feature | ModelBeat | OpenRouter | Martian | Not Diamond | Portkey |
|---|---|---|---|---|---|
| Single API key + credit billing | ✓ | ✓ | ~ | ✗ | ~ |
| 250+ models / 45+ providers | ✓ | ✓ | ~ | ~ | ✓ |
| Automatic failover + load balance | ✓ | ~ | ✓ | ✗ | ✓ |
| Per-request cost audit logs | ✓ | ~ | ~ | ✗ | ✓ |
| Cost-tier routing (L1) | ✓ | ~ | ✓ | ✓ | ~ |
| Multimodal routing | ✓ | ~ | ✗ | ✗ | ✓ |
| Built-in guardrails | ✓ | ✗ | ✗ | ✗ | ✓ |
| Quality prediction routing (L3) | Roadmap | ✗ | ✓ | ✓ | ✗ |
| Transparent pricing | ✓ | ✓ | ✗ | ✗ | ~ |
✓ = available · ~ = partial / platform tier · ✗ = not available · Roadmap = planned
We charge base provider cost plus a small, visible margin. You always know what you're paying and why. No hidden 20–50% markups.
in LLM spend vs. routing all traffic to frontier models — and counting.
Typical traffic distribution. Hard tasks still go to frontier models — that's the honest routing story.
We're onboarding teams in batches. Engineers at 300+ companies are already on the list. Get early access, locked-in pricing, and direct input on the roadmap.
No spam. No sales calls. Just a ping when your spot opens.