W44 Free & Discounted AI APIs (Oct 24-30, 2026): Tencent Token Plan Promo Ends Today 12:00, Baidu Qianfan From CNY 4.9, Volcengine 2.5x Off, and DeepSeek Search Bills From Nov 1
Week 44 free and discounted AI API roundup for October 24-30, 2026. Tencent Cloud Token Plan's two limited-time promotions expire at 12:00 Beijing time on October 30 with Hy3 at 50% off (CNY 0.50/2.00), GLM-5.3-Flash at 70% off (CNY 0.56/1.96) and Kimi K3 at 75% off (CNY 15/75). Baidu Qianfan Token Plan offers a 50% first purchase from CNY 4.9/month plus a 20% off-peak discount every night from 22:00 to 08:00. Volcengine Ark Coding Plan runs at 2.5 zhe for two months until November 8. Zhipu holds GLM-5.3-Flash at 50% off with Batch API at another half. Alibaba Bailian Coding Plan Pro drops to CNY 39.90 for the first month with over 100 million free tokens for new users. The deadline running the other way: DeepSeek Search and Fetch turn from free to USD 0.75 per 1,000 successful requests on November 1, which hits every agent workload using server-side tools.
Week 44's discount calendar is unusually dense, and it has two clocks running in opposite directions. On one side, a cluster of Chinese platform promotions is expiring within hours of this article being published - Tencent Cloud's Token Plan promotions close at 12:00 Beijing time on October 30, and several of the discounted per-token rates they offered are cheaper than the vendor's own list price, which does not happen often. On the other side, a free service many agent builders depend on becomes billable on November 1. This guide walks through every currently active discount, gives the exact deadlines, and explains which ones are actually worth restructuring your traffic around.
Tencent Cloud Token Plan: everything expires today at 12:00
Two concurrent promotions run on Tencent Cloud's Token Plan individual edition, and both close at 12:00 Beijing time on October 30, 2026. Activity 1 (October 1 to October 30, 12:00) covers Auto (tc-code-latest, a flat CNY 0.55 across input, output and cache), Kimi K3 at 75% off (CNY 15 input / CNY 75 output / CNY 1.5 cached), Kimi K2.7 Code at 80% off (CNY 5.2 / CNY 21.6 / CNY 1.04), GLM-5.3 at 55% off (CNY 4.4 / CNY 15.4 / CNY 1.1) and GLM-5.3-Flash at 70% off (CNY 0.56 / CNY 1.96 / CNY 0.161). Activity 2 (September 25 to October 30, 12:00) adds Hy4 preview at 50% off (CNY 3 / CNY 9 / CNY 0.15). The Hy Token Plan tier separately discounts Hy3 (hy3-202608) at 50% off (CNY 0.50 / CNY 2.00 / CNY 0.125). One yuan converts to twenty credits, so CNY 0.50 per million tokens on Hy3 works out to 10 credits. The important nuance: the promotion only adjusts the credit deduction price for the listed models - package quotas, refresh rules and usage limits are untouched, and prices revert to the standard credit table automatically at the deadline without any application step. Note how aggressive this is: Hy3 at CNY 0.50 is half of Tencent's own standard Hy3 rate of CNY 1 per million tokens, which means the reseller is currently selling the model below the vendor's list price.
Baidu Qianfan Token Plan: CNY 4.9 first purchase plus nightly 20% off
Baidu Qianfan's Token Plan individual edition lists four tiers, each with a 50% first-purchase discount: Mini at CNY 4.9 per month with 1,400 credits (list CNY 9.9), Lite at CNY 19.9 with 6,600 credits (list CNY 40), Pro at CNY 99.9 with 45,000 credits (list CNY 200) and Max at CNY 299.9 with 165,000 credits (list CNY 600). Renewals inside the campaign window get a further 30% off, and there is no seven-day quota cap on any tier - a notable change, since the old Coding Plan structure had three layers of rate limiting that throttled peak-hour usage even within your paid quota. The more interesting mechanic is the off-peak discount: DeepSeek-V4.1-Flash, DeepSeek-V4-Pro-0813 and GLM-5.3 all bill at 2 zhe, i.e. 20% of list, every day between 22:00 and 08:00. That is an 80% reduction on three of the strongest domestic models for a ten-hour window each night. If you run batch processing, nightly report generation or any scheduled ETL-style workload, moving those jobs to the night window is worth more than any first-purchase discount. The plan also moved to a dual credit-based and token-based billing model, works with Cursor, Windsurf and Cline over both OpenAI and Anthropic protocols, and keeps its API keys isolated from pay-as-you-go keys so a misconfiguration cannot silently bill you at list price.
Volcengine Ark and Alibaba Bailian: two-month first-purchase plays
Volcengine Ark's official discount rules confirm the Coding Plan Lite and Pro 2.5-fold promotion runs from June 8 to November 8, 2026. Lite (list CNY 40) drops to CNY 9.9 and Pro (list CNY 200) to CNY 49.9 for the first two months, reverting to list price in month three. The eligibility is shared across new purchases, renewals and tier upgrades, so upgrading does not hand back a fresh allowance - each account gets exactly two discounted months regardless of how you spend them. Capacity is first come first served, and Lite stacks with the separate 9.5-fold referral promotion. The plan bundles DeepSeek-V4.1-Flash, the GLM-5.3 family, Doubao-Seed-Evolving and Kimi-K3.1-pro behind an Auto routing mode. Alibaba Cloud Bailian's Coding Plan Pro has a comparable shape: list price CNY 200 per month, new-customer first month CNY 39.90, with quota restocked daily at 09:30 Beijing time on a first come first served basis. Its quotas are unusual in being explicitly request-based rather than token-based - 6,000 requests per 5 hours, 45,000 per week and 90,000 per month - and Alibaba warns that a simple task burns roughly 5-10 calls while a complex one burns 10-30 or more, so heavy reasoning workloads can exhaust the quota far faster than the raw numbers suggest. The Double 11 campaign layered on top raised the new-user free allowance to over 100 million tokens valid for 90 days, with student certification adding a CNY 300 no-threshold coupon and first purchases from 45% of list.
Standing discounts, and the one deadline that raises your bill
Three discounts are not time-boxed in a way you need to watch. Zhipu's official pricing page still lists GLM-5.3-Flash at a limited-time 50% discount - CNY 0.40 input, CNY 1.40 output, CNY 0.115 cached - against a standard rate of CNY 0.80 / 2.80 / 0.23, with a 1M context window, native multimodal input and Batch API at another 50% off. The page publishes no end date, so treat the discounted figure as subject to change. MiniMax has made its M3 discount permanent below 512K input tokens at CNY 2.10 / 8.40, doubling to CNY 4.20 / 16.80 above that threshold, while M3.1-Flash sits at CNY 1.00 / 4.00 with CNY 0.10 cached reads - and one yuan buys twenty credits on Tencent. The deadline that works against you is DeepSeek's: Search and Fetch are free right up until October 31, after which both cost USD 0.75 per 1,000 successful requests. Because server-side tool calls made through DeepSeek Harness, Claude Code and Codex already route through those same two APIs, every agent workflow that calls a search or fetch tool will start accruing per-request charges from November 1. The economics here depend heavily on request volume: at a few thousand tool calls a month the cost is trivial, but at hundreds of thousands it becomes a line item worth moving to a different provider. Kimi adds a migration deadline rather than a discount - the K2 series closes to new signups on October 31 and is fully retired in December, and since K3 carries 1,048,576 tokens against K2's 262,144, anyone still on K2 should treat this as a capacity upgrade as well as a deadline.
Frequently Asked Questions
Which AI API discount expires soonest right now?
Tencent Cloud Token Plan's promotions, at 12:00 Beijing time on October 30, 2026. Both the Activity 1 set (Auto at CNY 0.55, Kimi K3 at 75% off, Kimi K2.7 Code at 80% off, GLM-5.3 at 55% off, GLM-5.3-Flash at 70% off) and the Activity 2 set (Hy4 preview at 50% off) close at the same moment. They revert to the standard credit table automatically with no application step, so there is nothing to do in advance other than top up before noon.
What is the cheapest per-token API price right now?
Among domestic providers, Zhipu's GLM-5.3-Flash at its limited-time 50% discount is CNY 0.40 input / CNY 1.40 output per million tokens, with CNY 0.115 cached - effectively half of its own standard rate. Tencent's Token Plan resale briefly goes lower still at CNY 0.50 for Hy3, but only until October 30 at noon. Globally, DeepSeek's deepseek-flash off-peak rate of USD 0.15 input / USD 0.60 output per million tokens is the floor, with cached reads at USD 0.003. Batch API on Zhipu halves the standard price again for non-real-time work.
How does the Baidu Qianfan off-peak discount work and who benefits most?
DeepSeek-V4.1-Flash, DeepSeek-V4-Pro-0813 and GLM-5.3 bill at 2 zhe - 20% of list - every day between 22:00 and 08:00 the next morning, cutting those calls by 80%. It benefits batch processing, nightly report generation, scheduled ETL and any other job that can be shifted into the ten-hour window without latency constraints. It does nothing for interactive workloads, and note that the discount applies per request based on when Tencent's server receives it.
What changes on November 1 for DeepSeek users?
The Search and Fetch APIs stop being free. From 00:00 Beijing time on November 1, 2026 both cost USD 0.75 per 1,000 successful requests. Because server-side tool calls made through DeepSeek Harness, Claude Code and Codex already hit these same APIs, agent workflows that call search or fetch tools will begin accruing per-request charges. Model inference pricing is unchanged - deepseek-flash stays at USD 0.15 / 0.003 / 0.60 off-peak and deepseek-v4-pro at USD 0.66 / 0.022 / 1.98, with peak rates exactly double.
Should I buy a Coding Plan or a Token Plan?
Buy a Coding Plan if you only want an IDE or CLI assistant. Alibaba Bailian explicitly restricts Coding Plan keys to interactive use inside tools such as Claude Code, Cursor, Qoder and OpenCode, and treats API-style batch calling as abuse that can get the subscription suspended. Coding Plan quotas are also metered in model-call counts - 5-10 calls for a simple task, 10-30+ for a complex one - so reasoning-heavy work burns them quickly. Buy a Token Plan when your workload spans coding plus agents, RAG, data processing or batch work: Baidu Qianfan's Token Plan shares one quota across all scenarios, and Tencent's Token Plan covers both general and Hy agent use.
Is the GLM-5.3-Flash 50% discount permanent?
No. Zhipu's pricing page labels it a limited-time 50% discount and does not publish an end date, so the discounted CNY 0.40 / 1.40 / 0.115 rate is subject to change at any time. Plan around the standard CNY 0.80 / 2.80 / 0.23 rate for budget forecasts, and treat the discounted figure as upside. By contrast MiniMax's M3 discount below 512K input tokens is explicitly labelled permanent, so CNY 2.10 / 8.40 there is a safe baseline to build on.
Want more AI pricing comparisons?
Check our price changelog for daily updates on AI model pricing, new releases, and deprecations.