On 9 November, Alibaba Cloud raised the monthly quotas on the individual version of its Token Plan across the board. Monthly fees on the Lite, Essential and Standard tiers were left unchanged while quotas rose by 74 to 78 per cent. Pro was handled differently, with its quota lifted from 180,000 to 200,000 credits and its monthly fee cut from RMB 499 to RMB 312.
The cheapest Lite tier moved from 11,500 credits a month to 20,000, still at RMB 39. Essential went from 25,500 to 45,000 at RMB 79, and Standard from 45,000 to 80,000 at RMB 139. All three are more volume at the same price. Pro added only 20,000 credits but cut its fee by 38 per cent.
The arithmetic
Dividing monthly fee by monthly quota, the unit cost of a credit fell by more than 40 per cent on all four tiers. On Lite, RMB 39 bought 11,500 credits before, about RMB 0.0034 each. The same RMB 39 now buys 20,000, which puts the unit cost near RMB 0.002.
The change is not a new-customer offer. Existing users are included: from 9 November the new quotas apply, and the total allowance inside a current billing cycle is topped up to the new standard. Pro subscribers keep their current price for this cycle and move to the new price next cycle. Most platforms discount only for new registrations, and topping up existing users signals something different from acquisition spending.
Token Plan is the subscription service from Alibaba Cloud’s Bailian platform. It connects to Qwen office tools, the Qwen app, Qoder, Qwen Code and Claude Code, and a single credits pool covers different model calls and agent harness tooling across text and multimodal models. Image and video generation models draw from the same pool.
For an individual developer the practical gain is simpler billing. Instead of applying for separate API access per model and watching several bills, one credits pool covers text, code and multimodal, so a single top-up at the start of the month buys room to experiment. The heavier users of Claude Code and Qwen Code benefit most, because editing and debugging no longer have to be rationed.
Stacked on top of a shopping festival
The quota increase lands on top of the Double 11 promotion. Alibaba Cloud has the plan at a limited 50 per cent discount, with a first month at RMB 19.5 for Lite and RMB 39.5 for Essential. More volume at the same price, then half price on top of it, pushes the entry cost for an individual developer very low for the month.
Team pricing had already moved down. The cheapest standard seat is RMB 150 a month for 25,000 credits, the advanced seat RMB 550 for 100,000, and the top tier RMB 1,398 for 250,000. Individual and team lines are both moving down together, which makes the whole thing look less like a promotion and more like a round of concessions across the subscription system.
One caveat: individual quotas reset each billing month and unused credits do not roll over, so the increase rewards people who actually use the allowance inside the month rather than those who accumulate it.
A year of cuts, moved to subscriptions
Alibaba Cloud has been cutting all year. Bailian has reduced model API prices several times in 2026: qwen3-max in January, Qwen3-VL-Rerank text and multimodal input in August, and the speech model family in September. The quota increase extends that line, moving the battlefield from per-token billing to monthly subscription.
China’s model price war has separated into layers by 2026. DeepSeek cut output prices permanently by 75 per cent, while Zhipu’s GLM went the other way with a rise of more than 80 per cent and still saw call volume rise fourfold. Domestic inference costs have fallen by more than 90 per cent in two years, which is what gave Alibaba Cloud the room to add volume on the subscription side.
It picked volume rather than price, and it counted existing users in the move, which points at retention inside its agent tools rather than pure acquisition. Qoder, Qwen Code and Claude Code all connect to Token Plan, and a larger allowance makes users more willing to run extra rounds of trial and error inside those tools.
The wider signal is the scissors gap. China is adding volume while overseas AI coding tools raise prices. Cursor repriced after its October acquisition and removed the USD 20 Pro tier, Claude Code pushes heavy users toward USD 100 to 200, and Windsurf went from USD 15 to USD 20. Some overseas reviews have called 2026 the end of the twenty-dollar AI coding era, on the argument that frontier inference costs cannot be suppressed and vendors have stopped subsidising acquisition the way they did in 2024 and 2025.
That is the split to watch. One market is competing on unit cost and the other on willingness to pay, and the two are pricing the same capability.
Editor’s note: This is an adapted translation of the original OFweek report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://www.ofweek.com/ai/2026-11/ART-201718-8120-30710999.html.