Long-Context Chinese Models: What 1M Tokens Actually Costs

Long context is usually sold as a feature, not priced as one. The number that matters is simple: what does one pass over a full context window cost? That is just the input price per million tokens multiplied by the window size. Here is how the Chinese catalog looks.

The tiers

Every model with a long window

Three things the table shows

1. Context size and price are independent

Doubao-Seed-2.0-Lite has a 256K window at $0.075 per 1M input, while Kimi K2.7 Code HighSpeed has the same 256K window at $2.073 — a 28x difference for the same capacity. If your bottleneck is fitting the document, the cheap tier is enough. If your bottleneck is reasoning quality over that document, the expensive tier is the point.

2. The cheapest 1M window is not a flagship

GLM-5.3-Flash offers a 1M window at $0.08 per 1M input — cheaper per token than most 128K models. A full pass over a million tokens costs eight cents. Kimi K3 charges $2.90 for the same pass, which buys reasoning depth rather than capacity.

3. 4M exists, and it is not the most expensive

MiniMax Text-01 lists a 4M window at $0.14 per 1M input, so a full 4M pass is $0.56 — less than a single full pass on Kimi K3's 1M window. Very large windows are a throughput feature, not a premium tier.

What a full pass buys you

• ~128K tokens — a 300-page book, a mid-sized repository's relevant files, or a long support thread
• ~1M tokens — a large codebase, a year of meeting transcripts, or a full documentation set
• ~4M tokens — multiple books, an entire small product's source history

In practice most "long context" requests fill 10-20% of the window, which cuts the numbers above by 5-10x. The full-pass figure is the ceiling, not the bill.

Calling a long-context model

curl https://aiapi-pro.com/v1/chat/completions \
  -H "Authorization: Bearer $NOVAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"glm-5.3-flash","messages":[{"role":"user","content":""}]}'

Model ids are the same across every provider; switch by changing one string.

FAQ

What is the cheapest 1M-context model?

GLM-5.3-Flash at $0.08 per 1M input tokens, so one full-context pass costs about $0.08. Kimi K3 also offers 1M context but at $2.90 per 1M input, roughly 36x more for the same volume.

Which model has the largest context window?

MiniMax Text-01 lists a 4M-token context window, the largest in the catalog. A full 4M-token pass costs about $0.56 at its $0.14 per 1M input price.

Does a bigger context window always cost more?

No. Context size and price are independent. Doubao-Seed-2.0-Lite is the cheapest 256K model at $0.075 per 1M input, while Kimi K2.7 Code HighSpeed costs $2.073 per 1M input at the same 256K window.

Are these prices per request or per token?

Per token. You pay for the tokens you send and receive, not for the size of the window. A 10K-token request on a 1M-context model costs 1% of a full-context pass.

Context windows and prices reflect the live catalog on 2026-10-22 and are published machine-readably at /china-ai-leaderboard.json. "Full pass" is a derived figure: input price per 1M tokens × window size.