GLM Coding Plan vs Kimi, MiniMax and Qwen: Prices Compared

The GLM Coding Plan is the cheapest published subscription from a Chinese AI lab as of 4 October 2026: Lite starts at $18 per month, against $22 for MiniMax's M Plan Go tier, $50 for Alibaba Cloud's Coding Plan Pro and $20 for Claude Pro. Kimi Code is bundled into Kimi membership, and DeepSeek sells no subscription, only pay-as-you-go tokens. Every one of them offers an Anthropic-compatible endpoint for Claude Code, but Anthropic does not support non-Claude models there.

How much does each Chinese AI coding plan cost?

Prices and limits below were read from each vendor's own pages on 4 October 2026 and can change. Limits are quoted as the vendor states them.

Option	Price	Models	Limits as stated
GLM Coding Plan	Lite from $18/month; Pro and Max prices not readable on the docs	GLM-5.3, GLM-5.3-Flash	Credits per 5 hours / week: Lite 2,000 / 10,000, Pro 12,000 / 60,000, Max 28,000 / 140,000
Kimi Code	About $19/month at the entry tier (Moderato, now Plus), per Kimi's July 2026 explainer	K3, K2.8 Preview	5-hour rolling window plus a monthly total; paid Extra Usage on top
MiniMax M Plan	Go $22, Explore $55, Build $132 per month	M3.1 Flash Preview (text)	Explore 3× and Build 7.5× the usage of Go; 5-hour and weekly windows
Alibaba Coding Plan Pro	$50/month; Lite closed to new buyers	Qwen3.7-Plus, Qwen3.6-Plus, Kimi K2.5, GLM-5, MiniMax-M2.5 and more	6,000 requests per 5 hours, 45,000 per week, 90,000 per month
DeepSeek API	No plan; per 1M tokens, Flash $0.15–0.30 in, $0.60–1.20 out; Pro $0.66–1.32 in, $1.98–3.96 out	V4.1 Flash, V4-Pro	Concurrency 2,500 (Flash) and 500 (Pro)
Claude Pro	$20/month, $17 on annual billing	Claude models, Claude Code included	"Usage limits apply"; Max from $100 gives 5× or 20× Pro usage
Claude API (pricing)	Opus 5.5 $4 in, $20 out; Sonnet 5.5 $2 in, $10 out per 1M tokens	Claude models	Pay as you go

Two caveats. Kimi renamed its tiers (Go, Plus, Pro, Max) and its docs say pricing is unchanged, but we could not read the new USD table. MiniMax's older Token Plan, at the same prices, is closed to new purchases, and its launch discount of 50% off the first month runs through 14 October.

Which plan gives the most coding for the money?

The vendors measure usage in different units, so a fair comparison needs arithmetic.

• Alibaba: $50 ÷ 90,000 requests is about $0.00056 per model call if you use the full monthly cap.
• Z.ai: its credit formula for GLM-5.3 is (input × 6.9 + cached input × 1.7 + output × 24) ÷ 10,000. A session of 3M uncached input and 150K output tokens costs (20.7M + 3.6M) ÷ 10,000 = 2,430 credits at the peak rate, more than Lite's 2,000-credit 5-hour window. At a 95% cache hit rate it costs 948 credits, so Lite's 10,000 weekly credits cover about 10 such sessions at peak rates and about 21 off-peak, where credits are charged at 50%. Peak is Monday to Friday, 14:00–18:00 Singapore time. Z.ai also runs an all-day off-peak promotion from 25 September to 7 October 2026.
• DeepSeek: the same 3M-plus-150K session costs $1.08 on V4.1 Flash at peak rates and $0.54 off-peak, with no cap beyond concurrency.

Z.ai states that using off-peak discounts saves up to 92% against pay-as-you-go GLM-5.3 prices. That is the vendor's claim. Claude Pro and the Claude API cannot be converted the same way, because Anthropic gives no numeric allowance for Pro.

Do they work with Claude Code?

Yes, technically, through an Anthropic-compatible endpoint and the ANTHROPIC_BASE_URL variable. Anthropic's gateway documentation says it does not support routing Claude Code to non-Claude models, so you are on the vendors' support, not Anthropic's.

Provider	Anthropic-compatible base URL	Official guide
Z.ai	https://api.z.ai/api/anthropic	Z.ai Claude Code guide
Kimi Code	https://api.kimi.ai/coding/	Kimi Claude Code guide
MiniMax	https://api.minimax.io/anthropic	MiniMax Claude Code guide
Alibaba	https://coding-intl.dashscope.aliyuncs.com/apps/anthropic	Coding Plan docs, linked in the table above
DeepSeek	https://api.deepseek.com/anthropic	DeepSeek Claude Code guide

Z.ai's guide for the GLM Coding Plan puts this in ~/.claude/settings.json. Replace the key and restart Claude Code.

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "your_zai_api_key",
    "ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm-5.3-flash[1m]",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "glm-5.3[1m]",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "glm-5.3[1m]",
    "CLAUDE_CODE_AUTO_COMPACT_WINDOW": "1000000",
    "API_TIMEOUT_MS": "3000000"
  }
}

Each vendor's guide differs: Kimi's has a script that clears old settings and skips the Anthropic login, MiniMax's offers npx -y mmx-cli@latest agent setup, and DeepSeek's maps claude-opus names to V4-Pro and claude-sonnet or claude-haiku names to Flash. Our post Claude Code with any model covers the general mechanics.

Two setup traps recur in the vendors' own guides. Old ANTHROPIC_* variables left in your shell or in settings.json override the new ones, which MiniMax and Kimi both warn about, so clear them first. Then start claude and run /status and /model to confirm the session really points at the new provider before you trust any cost figure.

What terms and privacy rules are worth reading first?

• Z.ai: the plan is limited to officially supported tools, accounts cannot be shared, subscriptions are non-refundable and auto-renew, and cancellation is due at least 3 days before billing. Its data processing terms say customer data is generally processed in Singapore and that it does not store content you send.
• Kimi: interactive use only. The community guidelines forbid reselling access, non-interactive automation and altering the client's identity (User-Agent), which can suspend benefits.
• MiniMax: M Plan is for individual, interactive use, with pay-as-you-go recommended for production. Unused usage does not carry over.
• Alibaba: interactive use only, no scripts or application backends, no sharing, non-refundable, with limited slots restocked daily at 00:00 UTC+8.
• DeepSeek: its privacy policy says it collects, processes and stores personal data in the People's Republic of China.

Which payment methods and taxes apply?

Payment details are thin on the vendors' docs. Z.ai says fees go to a linked payment method such as a bank card or PayPal. Kimi and Claude state prices exclude tax, which is added at checkout by region. For Kimi's plan page, MiniMax, Alibaba and DeepSeek, the accepted payment methods are not published in the pages we read, so check the checkout before you commit. Z.ai and Alibaba plans are non-refundable, and Kimi's Extra Usage is generally non-refundable.

Which plan fits which kind of developer?

Situation	Reasonable option	Why
Lowest fixed monthly cost	GLM Coding Plan Lite	$18 per month, the lowest published entry price
Highest published request cap	Alibaba Coding Plan Pro	90,000 requests per month for $50, with Qwen, GLM, Kimi and MiniMax models
Heavy daily agent use on one bill	MiniMax Build	7.5× the usage of Go, at $132
Scripts, CI or a product backend	Pay-as-you-go API	The plans above state interactive use only
Staying on Claude models	Claude Pro or Max	Claude Code is included, and Anthropic supports it

How do you check a plan's limits before you pay?

1. Find the unit: credits, requests or "agents". They are not interchangeable.
2. Convert one of your real sessions into that unit, as we did for Z.ai above.
3. Check peak hours. Z.ai's peak is Monday to Friday, 14:00–18:00 Singapore time, which is 09:00–13:00 in Bucharest while summer time lasts (UTC+3 until 25 October 2026). DeepSeek's peak is 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday.
4. Read the refund and cancellation rules. Z.ai and Alibaba are non-refundable.
5. Start with a monthly term, not an annual one, until the limits fit your habits.

Can My Claude Code route Claude Code tiers to these providers?

Yes. My Claude Code, our local gateway, maps Claude Code's Opus, Sonnet and Haiku tiers to different providers with fallback chains and logs cost per request. That is one way to send Opus-class work to one provider and Haiku-class work to a cheaper one. It is subject to each provider's terms. Kimi forbids altering client identity, Z.ai limits its plan to supported tools, and Alibaba allows interactive use only, so a proxy may breach a plan even when the protocol works. Pay-as-you-go keys carry fewer such conditions. Read the terms first.

For model choices behind these plans, see the best open-weight coding models, and for the subscription-versus-API question on Claude itself see Claude subscription vs API.

How Danube Labs can help

Danube Labs helps teams pick the cheapest model that passes their own tests and wires it into their tooling. If you want a second opinion on plan limits, terms or routing, book a free intro call.