China Coding API Roundup 2026: Ark vs Bailian vs MiniMax vs Zhipu vs DeepSeek

China’s AI coding API subscription market exploded in 2026 — five major providers now offer fundamentally different product shapes: from Ark’s daily first-month flash sales (¥9.9, renew at ¥40) to Zhipu’s 30%-off annual deals, from MiniMax’s full-modal bundle to DeepSeek’s pure pay-as-you-go. Which one to buy? This is a roundup for developers actually making the decision.

TL;DR — Decide in 4 lines

• Cheapest first-month trial: Ark Coding Plan Lite first month ¥9.9 (renews at ¥40)
• Want stable GLM-5.1 / maximize annual discount: Zhipu Coding Plan Pro ¥149 (¥104/mo annual)
• Heavy multimodal (audio/video included): MiniMax Token Plan from ¥29
• Variable usage / want direct V4-Pro flagship: DeepSeek API (pay-per-token, V4-Pro currently 75% off)

For team scenarios or heavy Claude Code users, read on.

One Big Table

Pick by Scenario

🪙 Scenario A: Budget < ¥30, just want to experience it

Verdict: Ark Lite first month ¥9.9 (renews at ¥40)

Daily flash sale at 10:30 — new users get ¥9.9 first month, down to ¥8.9 with referral coupon. Fits daily chat + light agent tasks (1,200 calls per 5h). Second month onwards is ¥40 — if you need to stay under ¥30/mo long-term, switch to MiniMax ¥29 full-modal after the first month.

🦀 Scenario B: Claude Code with domestic models

Verdict: Ark (multi-model) / Zhipu (GLM) / Bailian (Qwen) / DeepSeek (V4-Pro flagship) — all four work

All four support Anthropic-protocol endpoints. Configure Claude Code with two env vars:

# Ark
export ANTHROPIC_BASE_URL="https://ark.cn-beijing.volces.com/api/coding/anthropic"
# Zhipu
export ANTHROPIC_BASE_URL="https://open.bigmodel.cn/api/anthropic"
# Bailian
export ANTHROPIC_BASE_URL="https://coding.dashscope.aliyuncs.com/apps/anthropic"
# DeepSeek (official Anthropic-compatible endpoint)
export ANTHROPIC_BASE_URL="https://api.deepseek.com/anthropic"
export ANTHROPIC_AUTH_TOKEN="<your-deepseek-key>"

Picker: want Auto model routing → Ark; want GLM-5.1 flagship → Zhipu; want full Qwen Coder lineup → Bailian; want V4-Pro flagship pay-as-you-go → DeepSeek (V4-Pro 75% off until 5/31).

📊 Scenario C: Variable usage, don’t want subscription lock-in

Verdict: DeepSeek API

The only non-subscription option — pure token-based pricing. V4-Pro currently 75% off (until 2026-05-31), 80.6% SWE-bench score — the cheapest flagship token rate available. Officially supports both OpenAI and Anthropic protocols — use the OpenAI endpoint with Cline / Aider / Cursor, or set ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic for native Claude Code support, no third-party wrapper required.

🎨 Scenario D: Heavy multimodal / all-modal Agent

Verdict: MiniMax from ¥29 or Ark Agent Plan

MiniMax Token Plan bundles text/image/video/audio/music — five modalities at a ¥29 entry. Ark Agent Plan from ¥40 includes Doubao-Seedance 2.0 (video) + Seedream 5.0 lite (image) + Harness toolchain (web search, memory); ¥200+ tier adds the 24/7 AI Companion.

⚠️ Agent Plan applies a 7.5× multiplier for long context (>128k) — long-running tasks burn quota fast.

👥 Scenario E: Team with multiple seats

Verdict: Bailian Token Plan Team Edition from ¥198/mo

One of the few Chinese products with explicit per-seat billing. Ark supports sub-accounts but doesn’t bill per seat. Others are individual-focused.

💰 Scenario F: Annual subscription for max discount

Verdict: Zhipu GLM Coding Plan

Zhipu has the most aggressive discount: 10% off quarterly / 30% off annually. Pro monthly ¥149, annual drops to ¥104/mo. Catch: only worthwhile if you’re already committed to GLM models — the savings only matter if you lock into a single family.

⚠️ Zhipu repriced 2026-02 with 30%+ increase and removed first-buy discounts. Entry cost is no longer as low as before.

Deep Dive: All Five

1. Volcengine Ark (Coding Plan / Agent Plan)

Position: Lowest entry price in China + broadest client support.

Strengths: Auto mode (routes by task complexity), 11+ client support including native Anthropic protocol for Claude Code, transparent daily 10:30 flash sale.

Watch out: 5h sliding window can throttle bursty workloads. Agent Plan tokens cost 2–2.5× more than Coding Plan — not recommended for pure text coding.

See Ark Coding Plan complete guide and Agent Plan explainer.

2. Alibaba Bailian (Coding Plan / Token Plan)

Position: Most models (8 including full Qwen lineup) + explicit team multi-seat.

Strengths: Only Chinese plan bundling Qwen Coder + Qwen Plus + GLM + Kimi + MiniMax. Team edition with per-seat billing — best for enterprise. Monthly quota mechanism — front-loadable.

Watch out: Lite discontinued (sold out March 2026, renewals stopped April). Entry now ¥200. No Auto mode — must specify model IDs manually.

See Bailian Coding Plan complete guide.

3. MiniMax Token Plan

Position: Lowest entry for full-modal subscription (audio/music included).

• Entry ¥29/mo
• Covers text / code / image / video / audio / music
• Best fit for multimedia teams, voice/music app builders

Watch out: Pure coding isn’t its core strength — coding ability lags Ark / Bailian / Zhipu somewhat.

See MiniMax Token Plan explainer.

4. Zhipu GLM Coding Plan

Position: GLM-5.1 flagship + best annual discount for committed users.

Strengths: GLM-5.1 ranks at the top of Chinese coding benchmarks. 30% annual discount is the most aggressive among peers. Anthropic protocol supported — Claude Code works out-of-box.

Watch out: Repriced 2026-02 with 30%+ increase, no first-buy discount. Only GLM-family models — no multi-vendor rotation.

5. DeepSeek API (pay-as-you-go, outlier)

Position: Only non-subscription option, only official direct channel for V4-Pro / R1.

• V4-Flash: $0.14/M input, 82.6% HumanEval
• V4-Pro: 80.6% SWE-bench, Codeforces #1 (3,206) — currently 75% off, ends 2026-05-31
• Web Chat fully free

Strengths: True usage-based pricing, no subscription lock. V4-Pro discounted is the cheapest flagship token rate in China. Direct developer access — no middlemen.

Watch out:

• Officially supports OpenAI + Anthropic dual protocols — Claude Code uses ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic natively, no third-party wrapper needed
• Steady users likely save more with subscription plans
• Web tier suffices for individual light usage

Decision Tree (30 seconds)

What's your monthly budget?
├─ < ¥20 → Ark Lite first month ¥9.9 (renews at ¥40)
├─ ¥30~¥100
│   ├─ Want full-modal → MiniMax ¥29
│   └─ Want GLM-5.1 → Zhipu Lite ¥49
├─ ¥100~¥300
│   ├─ Need team multi-seat → Bailian Team ¥198
│   ├─ Maximize annual discount → Zhipu Pro annual ¥104/mo
│   ├─ Solo heavy coding → Ark Coding Pro ¥200 (first month ¥49.9) or Agent Plan Small ¥40
│   └─ Multi-model rotation → Bailian Pro ¥200
├─ > ¥300 (heavy Agent)
│   ├─ Multimodal + Harness → Ark Agent Plan Medium ¥200
│   ├─ Team scale-up → Bailian Team upgrade
│   └─ Flexible usage + flagship → DeepSeek V4-Pro (pay-as-you-go)
└─ Very unstable usage → DeepSeek pay-as-you-go

My Final Recommendation

If you want a single default recommendation: start with Ark Lite first month ¥9.9, measure your real consumption, then decide whether to upgrade or migrate. ¥10 buys you data — and solves the “which tier do I actually need” decision paralysis for 80% of users. Just remember the second month onwards is ¥40 — budget against that.

After one month, if you find yourself:

• Constantly hitting the 5h window → upgrade to Coding Pro or Agent Plan
• Wanting GLM flagship + can afford annual → migrate to Zhipu Pro annual
• Scaling team → switch to Bailian Team
• Usage too variable → switch to DeepSeek pay-as-you-go

Related

• Coding Plan landing page — all 5 plans + recommended bundles
• Bailian vs Volcengine Ark Coding Plan
• Cline + Ark setup guide
• Cursor in China complete guide
• GitHub Copilot June billing change

Verified through 2026-05-19. All prices, quotas, and promotions per official channels — Chinese API pricing moves quickly, double-check before subscribing.