Solutions
Cheap reasoning model API
For frontier reasoning models without frontier bills, Somnus gives you 12 models on one OpenAI-compatible key. Usage is billed at list price but every top-up is credited 2× to 2.5×, so GLM 5.3 — the cheapest option here — effectively costs $0.7 / $2.2 per 1M input/output tokens.
Choosing a cheap reasoning model api
Reasoning models think before they answer, and those hidden thinking tokens are billed as output. That makes the output price the number that matters most for reasoning workloads.
Limit reasoning effort where the API allows it, and reserve top-tier models for the problems that genuinely need multi-step thinking.
Cheap reasoning model API: price ranking
| # | Model | Maker | List (in / out) | Effective with Somnus |
|---|---|---|---|---|
| 1 | GLM 5.3 | Zhipu | $1.4 / $4.4 | $0.7 / $2.2 |
| 2 | GPT-5.6 Luna | OpenAI | $1 / $6 | $0.5 / $3 |
| 3 | Qwen 3.8 Max | Alibaba | $1.65 / $4.95 | $0.83 / $2.48 |
| 4 | Grok 4.7 | xAI | $2 / $6 | $1 / $3 |
| 5 | Claude Sonnet 5 | Anthropic | $2 / $10 | $1 / $5 |
| 6 | Gemini 3.1 Pro | $2 / $12 | $1 / $6 | |
| 7 | Kimi K3 | Moonshot | $3 / $15 | $1.5 / $7.5 |
| 8 | Claude Opus 5.5 | Anthropic | $4 / $20 | $2 / $10 |
Ranked by blended price (3 input : 1 output tokens).
Why Somnus for cheap reasoning model api
- Somnus buys model access in bulk and passes the saving on as extra credit.
- Pay-as-you-go. No monthly caps or throttling — spend your balance however you like.
- $5 free credit on sign-up, no card needed.
- Switch models by changing one string — same key, same balance.
Point the OpenAI SDK at Somnus
import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.SOMNUS_API_KEY,
baseURL: "https://gateway-production-c837.up.railway.app/v1",
});
const resp = await client.chat.completions.create({
model: "claude-sonnet-5",
messages: [
{"role": "user", "content": "Cheap reasoning model API"}
],
});
console.log(resp.choices[0].message.content);Pick the model name from your dashboard.
| Claude Sonnet 5 | GPT-5.6 Sol | |
|---|---|---|
| Input / 1M | $2 | $5 |
| Output / 1M | $10 | $30 |
| This request | $0.0120 | $0.0340 |
| Blended / 1M (3:1) | $4.00 | $11.25 |
Claude Sonnet 5 is 2.8× cheaper for this request.
Questions
What is the cheap reasoning model api?
Among frontier reasoning models without frontier bills, GLM 5.3 has the lowest list price on Somnus ($1.4 / $4.4), about $0.7 / $2.2 effective.
Is it really that cheap?
Yes: usage is charged at list price, and each $1 you top up becomes $2–$2.50 of credit.
Related
Browse more: Alternatives · Compare · Use cases · Developers · Solutions · Tools