Guides · dev · ai coding plan comparison
Coding plans compared: GLM's credits, and the plans we would not price
A coding plan is a bundle: a model, a cap, and a client. Comparing them by the monthly price alone is how a $18 plan looks like a win until the weekly credits end on Tuesday.
What is actually on a vendor page
GLM, from the overview and the migration notice (see the GLM plan guide for the tables):
- Lite, Pro, and Max exist.
- Each has a 5-hour credit cap and a weekly credit cap. Lite is 2,000 and 10,000.
- The migration notice lists standard prices of $18, $72, and $160. Confirm at checkout. We saw a conflicting secondary list.
DeepSeek’s API pricing page is token billing with peak and off-peak. There is no plan SKU on that page. If you want a subscription feeling, you will not get it there. You will get a balance that falls when the agent runs.
Kimi Code’s membership guide describes K3 inside a membership, with a 256k variant that consumes less quota than the 1M variant. It does not, in the text we used, publish a dollar price next to a credit definition we can put in a table. We are not filling one in from memory.
Qwen’s international path is Model Studio metered tokens, plus whatever trial credit the console shows today. We did not read a “coding plan” SKU we were willing to price.
Volcengine Ark and Bailian token-plan prices circulate in roundups. We did not open those consoles on 16 Oct 2026. They are omitted on purpose.
The pay-as-you-go alternative
| Model | Session | vs Opus | vs Sol | Cache price |
|---|---|---|---|---|
| glm-5.3-flash | $0.096 | 97% cheaper | 94% cheaper | published |
| deepseek-v4.1-flash | $0.066 | 98% cheaper | 95% cheaper | published |
| minimax-m3 | $0.324 | 89% cheaper | 78% cheaper | published |
| kimi-k3 | $1.35 | 54% cheaper | 8% cheaper | published |
| Anthropic: Claude Opus 5.5 | $2.94 | — | — | published |
| OpenAI: GPT-6.1 Sol | $1.47 | — | — | published |
The percentages in this table compare session cost. The price table's percentages compare blended token price. They answer different questions.
If that illustrative session is in the right order of magnitude for your day, multiply by days and compare with $18. If your agent loops, the weekly GLM cap is the constraint, not the 5-hour one. Kimi’s bar on this chart is the OpenRouter list price. The Kimi guide shows why the official output rate changes the story.
A practical order
- Spend a small topped-up balance on a flash model and keep the traces.
- If the traces are good enough, stay on tokens.
- If you want a cap you can explain to yourself, try GLM Lite and write down the day the weekly credits die.
- Do not buy the top tier because a benchmark thread said the flagship is “Claude class”. The comparison is the antidote.
Related: Claude Code setup, paying from abroad.
Questions
Which Chinese coding plan is cheapest?
We can only price GLM from the vendor docs we fetched. DeepSeek does not sell a coding plan on its API pricing page. Other subscriptions need a fresh look at checkout.
Can I buy these with a card outside China?
We did not complete a payment. The paying-from-abroad guide says what we could and could not verify.
Sources
- GLM Coding Plan overview, accessed 2026-10-16
- GLM Coding Plan migration notice, accessed 2026-10-16
- DeepSeek API pricing, accessed 2026-10-16
- Kimi Code membership guide, accessed 2026-10-16