Lumen gets better every day
One engine, three products, the same model ids forever. Watch it work below — then see the daily measurements that prove the trend.
Animated sample session: a request streams back a complete Lumen answer in real time.
- 3
- tiers, evaluated daily
- —
- Index calibrating
- Position-swapped
- judging, every duel
The trend is measured, not asserted.
A five-part quality score, refreshed by scheduled runs, published for every tracked model — not a number we picked once and kept repeating.
No version numbers to chase
New lanes are admitted, rated, and promoted as the market moves. Every improvement lands behind the exact model id you already integrated — automatically, continuously, forever.
The advertised context window today — a floor enforced in code. It only rises from here. Never regressed, never just a marketing figure.
Every claim is checked in production
Quality isn't asserted once. It's re-verified continuously, on real traffic, before anything ships.
Shadow duels
A sibling lane invisibly answers the same request, off your path. A family-disjoint judge compares both — position-swapped both ways, so order never decides the winner.
Zero retention
Only the anonymous win/loss score survives a duel. Never the prompt, the output, or who sent it.
Promotion gates
A lane earns a serving seat only after enough judged duels at or above the pool's median — shipped as a reviewed config change, rolled back automatically if live quality slips.
One engine. Four ways to spend on it.
Every tier runs the same router, the same lanes, the same certification. Tier separation — Flash ≤ Standard ≤ Pro ≤ Ultra — is checked continuously against production duels, never asserted once.
Glow
max_passes 2Thrift budget. Same engine, tighter deepen allowance.
sylphx/glowLumen
max_passes 3The default. Balanced deepen budget.
sylphx/lumenFlare
max_passes 4Higher willingness to spend time on hard turns.
sylphx/flareReady to build?
Get your API key in under a minute. No credit card required to start.
import OpenAI from "openai"
const client = new OpenAI({
apiKey: "sk-sx-...",
baseURL: "https://api.sylphx.ai/v1",
})
const response = await client.chat.completions.create({
model: "sylphx/lumen",
messages: [{ role: "user", content: "Hello!" }],
})
console.log(response.choices[0].message.content)