Individual edition launches: Coding Plan upgraded to Token Plan
Baidu Qianfan Token Plan Individual launched on 7/13, upgrading the old Coding Plan into a unified "one Token quota, multi-vendor models" pack. The official campaign page positions it as a personal AI compute traffic pack with four tiers from Mini to Max and a 50% off first-purchase flash sale (source).
The biggest change isn't just a new tier — the billing logic changed. The old Coding Plan billed per request with multi-layer rate limiting; the new plan uses unified Token deduction with no model multiplier, no input/output split, and no cache-hit distinction. Every model deducts by actual Token consumption, so the math is simpler and model choice is freer.
What actually changed
- Billing: from per-request with rate limits to unified Token deduction.
- Model pool: from ERNIE-only to ERNIE plus GLM, Kimi and DeepSeek multi-vendor.
- Rate limits: the old three-layer sliding-window limit is removed for stable in-quota calls.
Four-tier pricing comparison
The pricing comparison across the four tiers is straightforward: higher quota, higher entry price, with the first purchase starting as low as ¥4.9. The four tiers are (source):
| Tier | Monthly | 50% off first buy | Token quota | Positioning |
|---|---|---|---|---|
| Mini | ¥9.9 | ¥4.9 | 10M | Trial |
| Lite | ¥40 | ¥19.9 | 42M | Daily dev |
| Pro | ¥200 | ¥99.9 | 230M | Advanced coding (most popular) |
| Max | ¥600 | ¥299.9 | 700M | Heavy professional |
The pricing comparison conclusion: start with Mini or Lite to sample, or go straight to Pro for heavy use. The 50% off is a daily 10:00 limited-quota flash sale and renewals return to regular prices, so don't treat the first-purchase price as your long-term cost.
One quota across ERNIE, GLM, Kimi and DeepSeek
This is the core of the model-pool restructure: one Token quota lets you switch freely across ERNIE, GLM, Kimi and DeepSeek flagships without buying each separately. The pool listed on the campaign page includes (source):
| Vendor | Models | Notes |
|---|---|---|
| Baidu ERNIE | ERNIE 5.1 | Latest ERNIE, officially retiring 8/20 |
| Zhipu GLM | GLM-5.2 / GLM-5.1 / GLM-5 | Long-horizon tasks and agent engineering |
| Kimi | Kimi-K2.6 | General agent, coding and vision |
| DeepSeek | DeepSeek-V4-Pro / V4-Flash | Reasoning and coding flagships |
The point of this pricing comparison: you no longer need a separate subscription to try DeepSeek or GLM — one quota covers horizontal switching. Just note ERNIE 5.1 is officially marked for retirement on 8/20, so check that date before making it your primary model.
Pricing comparison vs pay-as-you-go — worth it?
Compared with Qianfan's pay-as-you-go API, Token Plan is worth it mainly if you're a high-frequency coding or agent user who wants a fixed monthly fee. The two are independent systems (source):
| Dimension | Token Plan Individual | ERNIE API pay-as-you-go |
|---|---|---|
| Billing | Fixed monthly + Token quota | Per-token postpaid |
| Models | ERNIE + GLM + Kimi + DeepSeek | Mostly ERNIE self-developed |
| Use case | Interactive coding tools | Backend services, automation |
| API key | Dedicated, coding tools only | Open API key |
The pricing comparison conclusion: if you run agents daily in Claude Code or Cursor, Token Plan's fixed fee plus multi-model switching is more convenient; if you only call the API occasionally or need backend batch jobs, pay-as-you-go fits better. The dedicated key must not be used for automation scripts or app backends.
API key, integration, quota and rate limits
Token Plan Individual has an isolated dedicated API key, fully separate from pay-as-you-go and the enterprise edition. Key points:
- The dedicated key plus a Base URL connects to Claude Code, Cursor, Cline, OpenCode, Qwen Code and other tools over OpenAI/Anthropic protocols (source).
- Quota deducts uniformly by Token; the console shows call counts, Token consumption and remaining balance.
- The old three-layer sliding-window rate limit is removed, so calls run stably within quota.
- Limited-time off-peak 20%: daily 21:00–08:00, DeepSeek-V4-Pro, GLM-5.2, DeepSeek-V4-Flash and DeepSeek-V4-Flash-0731 cost only 20% Token.
- Enterprise "more credits, same price" adds bonus credits, but the enterprise edition meters by credits, not Token.
Things to watch
-
The 50% off first purchase is limited and renewals return to regular price. The daily 10:00 flash sale (Mini ¥4.9 / Lite ¥19.9 / Pro ¥99.9 / Max ¥299.9) is for new users' first purchase; budget long-term at regular prices.
-
The dedicated key has a hard boundary. It's for interactive coding-tool use only — no automation scripts, app backends, or batch calls — and is not interchangeable with the pay-as-you-go API key.
-
ERNIE 5.1 is retiring. Officially marked for 8/20 retirement, so migrate early if your existing workflow depends on it.
-
Off-peak 20% has specified models and a time window. It only applies daily 21:00–08:00 to specific models such as DeepSeek-V4-Pro and GLM-5.2 — don't assume all models are discounted around the clock.