Blog

Cheapest LLM API in 2026: DeepSeek V4.1 Flash vs Qwen Flash vs GLM-5.3-Flash vs MiniMax vs MiMo vs Doubao Lite

Short answer

At official list prices on November 17, 2026, the budget tier of Chinese models (the Flash and Lite models) costs $0.03 to $0.15 per million input tokens and $0.11 to $0.60 per million output tokens — one to two orders of magnitude below flagship models.

  • Cheapest overall: Qwen3.7 Flash at ¥0.2 / ¥0.8 (about $0.03 / $0.11). 100M input + 10M output tokens a month comes to about $4.
  • The current-generation budget models are priced alike: Qwen3.8 Flash, Doubao Seed 2.1 Lite and GLM-5.3-Flash, all about $0.11 / $0.39, or about $15 for the same volume.
  • Cache-heavy agents: DeepSeek V4.1 Flash and Xiaomi MiMo V2.6 Flash. Cache hits cost $0.003 and about $0.003, 2% of the uncached input price. With 80% of input served from cache, MiMo comes to about $6 and DeepSeek about $9.
  • DeepSeek has peak pricing: the figures above are off-peak. Weekdays 01:00–04:00 and 06:00–10:00 UTC are peak hours, at double the price.

For reference, the same volume on Gemini 4 Hydrogen ($1.50 / $7.50) costs $225.

Get an API key and run these budget models on your own task

New accounts start with a $0 balance — top up before your first request. You can top up from $5, by card or with USDT, USDC or ETH. If a request returns 402 (400 in OpenAI format) with "wallet balance is insufficient", the balance is empty and your base URL and key are set up correctly. Top up and retry.

Side by side

Official list prices per million tokens; models priced in CNY are converted at about ¥7 = $1. DeepSeek prices come from DeepSeek's pricing page; the rest match the official prices shown on TeamoRouter's model pages, read on November 17, 2026. "Monthly cost" assumes 100M input and 10M output tokens.

Model Input Output Cache hit Context Monthly cost (no cache) Monthly cost (80% of input cached)
Qwen3.7 Flash ¥0.2 (≈$0.03) ¥0.8 (≈$0.11) ≈$0.006 1M $4.00 $2.17
Qwen3.8 Flash ¥0.8 (≈$0.11) ¥2.7 (≈$0.39) ≈$0.014 1M $15.29 $7.29
Doubao Seed 2.1 Lite ¥0.8 (≈$0.11) ¥2.7 (≈$0.39) ≈$0.023 1M $15.29 $7.97
GLM-5.3-Flash (Ox Alpha) ≈$0.11 ≈$0.40 ≈$0.03 1M $15.40 $8.68
MiMo V2.6 Flash ¥1 (≈$0.14) ¥2 (≈$0.29) ≈$0.003 1M $17.15 $5.95
DeepSeek V4.1 Flash $0.15 (peak $0.30) $0.60 (peak $1.20) $0.003 1M $21.00 $9.24
MiniMax M3.1 Flash $0.15 $0.60 $0.015 1M $21.00 $10.20
Reference: Gemini 4 Hydrogen $1.50 $7.50 $0.075 1M $225.00 $111.00

Which one to pick

Lowest price for simple jobs (classification, tagging, short rewrites, bulk translation): start with Qwen3.7 Flash, at roughly a quarter of what the other budget models cost for the same volume. It is the previous generation, so for anything harder compare it with Qwen3.8 Flash on your own data.

Choosing among the current-generation budget models: Qwen3.8 Flash, Doubao Seed 2.1 Lite and GLM-5.3-Flash cost almost the same, so the difference is quality and output length: Doubao Seed 2.1 Lite can return up to 256K tokens in one response, Qwen3.8 Flash up to 128K. When prices tie, run 50–100 real samples of your own and keep the one with the best pass rate.

Agents, long chats, and anything that resends the same large context: the cache-hit price dominates. DeepSeek V4.1 Flash and MiMo V2.6 Flash charge about $0.003 per million cached tokens, so with 80% cache hits they cost less than Qwen3.8 Flash. DeepSeek V4.1 Flash also accepts images, returns up to 384K tokens per response, and DeepSeek offers an Anthropic-format endpoint as well.

Batch jobs that can wait: DeepSeek doubles its price at peak, so schedule batch work outside weekday peak hours (01:00–04:00 and 06:00–10:00 UTC) to pay the off-peak rate.

When the budget tier isn't enough: the next step up is the flagships — DeepSeek V4 Pro, Kimi K3.1 Pro, Qwen4 Max. Scores and prices are in Gemini 4 vs Kimi, Qwen and DeepSeek.

Calling them with one key

All of these models work through the same OpenAI-compatible endpoint; switching is a one-line change to model. To compare them on your own task:

python
from openai import OpenAI

client = OpenAI(base_url="https://api.teamorouter.com/v1", api_key="YOUR_KEY")

for model in ["qwen3.7-flash", "qwen3.8-flash", "doubao-seed-2.1-lite", "glm-5.3-flash", "mimo-v2.6-flash", "deepseek-flash", "minimax-m3.1-flash"]:
    resp = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": "Classify this customer message as bug, feature request or question. Reply with the label only: ..."}],
    )
    print(model, resp.usage, resp.choices[0].message.content[:200])

The usage log in the console shows what each run cost. Compare cost per acceptable result, not price per token: a cheap model that needs two retries is no longer cheap.

How to pay

No subscription: top up your balance in the console from $5 and each request is billed by the tokens it uses. Cards are supported, and so are USDT, USDC and ETH. No Chinese phone number or Alipay account is needed for any of these models. One balance covers all of them, plus flagships such as DeepSeek V4 Pro, Kimi and Qwen4 Max, and GPT and Claude.

FAQ

What is the cheapest LLM API in 2026?

Among the models compared here, at official prices, Qwen3.7 Flash: ¥0.2 / ¥0.8 per million tokens (about $0.03 / $0.11). Of the current-generation models, Qwen3.8 Flash, Doubao Seed 2.1 Lite and GLM-5.3-Flash are close at about $0.11 / $0.39.

How much does DeepSeek V4.1 Flash cost?

Off-peak: $0.15 per million input tokens, $0.60 per million output tokens, $0.003 per million cached input tokens. Prices double on weekdays from 01:00 to 04:00 and 06:00 to 10:00 UTC. Requests to the old name deepseek-v4-flash are now served by V4.1 Flash and billed at the Flash price.

Which cheap model is best for agents?

Agents resend context every turn, so the cache-hit price matters most. DeepSeek V4.1 Flash and MiMo V2.6 Flash charge about $0.003 per million cached tokens and have the lowest monthly cost at 80% cache hits (apart from Qwen3.7 Flash).

Can I use Chinese models without a Chinese phone number?

Yes. On TeamoRouter you sign up with an email, top up by card or crypto, and call DeepSeek, Qwen, GLM, MiniMax, MiMo and Doubao with one key and one base URL.

Related

Ready to connect?Log in · top up · create an API key — three steps to start.