Coding plan usage limits compared
Compare six current coding subscriptions by allowance unit and reset window, then estimate how much of your own agent workload each allowance covers.
Compare coding plans by the allowance you can spend on your workload, then by price. A “credit”, model call and percentage of a shared pool count different things; a cheaper subscription can run out sooner if your agent makes long replies or many tool calls.
This comparison uses the third-party terms checked on September 29, 2026. The main tables cover six current subscription offers from five providers; legacy Kimi plans, DeepSeek’s pay-as-you-go API, Claude’s relative limits and our Coding Plan appear separately. Prices retain their published billing currencies and are a dated snapshot, not a current checkout quote.
Current coding subscriptions: allowance units
| Plan | Price | Unit | Formula public | Signup |
|---|---|---|---|---|
| Z.ai GLM Coding Plan — Lite/Pro/Max | $18/$80/$168 | Credits | Yes | Open |
| Moonshot Kimi Code — Plus/Pro/Max/Ultra | $19/$39/$99/$199 | Shared pool % | No | Open |
| MiniMax Token Plan — Plus/Max/Ultra | $22/$55/$132 | Usage bar | No | Open |
| Alibaba Coding Plan — Pro | ¥200 | Model calls | Partial | Rationed, Lite closed |
| Alibaba Token Plan, Personal — Lite/Essential/Standard/Pro | ¥60/¥120/¥180/¥600 | Credits | No | Open |
| OpenCode Go — Go/Go Plus | $10/$40 | List-price dollars | Partial | Open |
Usage windows and resets
| Plan | 5‑hour | Weekly | Monthly | Reset |
|---|---|---|---|---|
| Z.ai GLM Coding Plan | 2,000/12,000/28,000 | 10,000/60,000/140,000 | Not stated | Rolling + fixed |
| Moonshot Kimi Code | Named, no figure | Removed | Shared pool, then frozen | Rolling |
| MiniMax Token Plan | Named, no figure | Named, no figure | No carryover | 5‑hour rolling |
| Alibaba Coding Plan | 6,000 requests | 45,000 requests | 90,000 requests | Rolling + fixed |
| Alibaba Token Plan, Personal | Not stated | Removed | 11,500/25,500/45,000/180,000 | Fixed |
| OpenCode Go | 20% of monthly | 50% of monthly | Per-model dollar limits | Not stated |
Reference plans and alternative billing
| Plan | Price | Unit | Formula public | Signup |
|---|---|---|---|---|
| Moonshot Kimi Code, legacy — CNY Andante/Moderato/Allegretto/Allegro | ¥49/¥99/¥199/¥699 | Shared pool % | No | Existing members |
| DeepSeek | Pay per token | Tokens | Per-token rates | Open |
| Anthropic Claude Pro/Max — reference subscription | $20/$100/$200 | Relative multiple | No | Open |
| Standard Thinking Coding Plan | $149/week, $549 per 30-day period | No allowance | Not applicable | On request |
| Plan | 5‑hour | Weekly | Monthly | Reset |
|---|---|---|---|---|
| Moonshot Kimi Code, legacy | Named, no figure | Named, no figure | Shared pool, then frozen | Rolling + fixed |
| DeepSeek | No window | No window | No window | None |
| Anthropic Claude Pro/Max | Named, no figure | Named, no figure | Not stated | Rolling + fixed |
| Standard Thinking Coding Plan | None | None | None | Terms |
Conditions behind the numbers
- Z.ai. A credit weights input, cached input and output tokens by model, then divides the sum by 10,000 — for GLM‑5.3, 6.9, 1.7 and 24. Documented off-peak adjustments also affect consumption.
- Kimi Code. New members of the plans released in September have no weekly quota, only the rolling 5‑hour window, and the 1M-context model consumes about twice the quota of the 256K one.
- Kimi Code, legacy. Membership features share one token-based credit pool; unused quota does not roll over.
- MiniMax. The monthly quota shows as “a usage bar in the console” and deducts at endpoint pricing; tiers are sized as 3-4, 4-5 and 6-7 agents. No peak policy is published.
- Alibaba Coding Plan. Quota counts model calls, not user turns: simple tasks typically use 5–10 calls, complex ones 10–30 or more. Under high platform load or concentrated use Alibaba may queue, throttle or briefly interrupt an account’s requests.
- Alibaba Token Plan. A Credit is shared across text, image, speech and video. The weekly quota was removed on September 22, 2026 for a monthly one on a 30‑day cycle from the subscription date; unused Credits expire, concurrency is 1–2 to 6–8 agents, and no peak policy appears.
- OpenCode Go. The budget is a monthly dollar limit per model rather than one pool: Kimi K3 $15, GLM‑5.3 $15 and GLM‑5.3‑Flash $60 on Go, and $60, $120 and $180 on Go Plus. DeepSeek’s peak rates pass through at double the off-peak rate, without its holiday exemption, and the limits may change with early usage.
- DeepSeek. Token charges are deducted from a balance, with model-specific peak and off-peak rates.
- Claude Pro / Max. No absolute unit is published; what a session allows varies with message, attachment and conversation length and the model or feature used. Peak is priority access for Pro “during high-traffic periods”, with a reserved clause allowing other limits at Anthropic’s discretion.
- Standard Thinking. Coding Plan is available on request. There is no 5-hour, weekly or monthly limit; requests may be slower at peak hours, with a baseline kept.
Estimate the allowance for one coding task
Record input, cached input and output tokens for a representative task, including subagents, retries and parallel calls. Also count model calls: one user turn can trigger several, so a request allowance cannot be divided by your number of chat messages.
For a plan with published token multipliers, apply them to that trace. Z.ai’s GLM‑5.3 formula is (input × 6.9 + cached input × 1.7 + output × 24) / 10,000; use uncached input for the first term so cached tokens are not counted twice. A trace with 10,000 uncached input, 20,000 cached input and 2,000 output tokens consumes 15.1 credits before any documented off-peak adjustment. This is a calculation example, not measured task usage.
For a call-based plan, divide each window’s allowance by calls per task. For a dollar budget, price the trace at the specified model’s list rates. If the provider publishes only a percentage or usage bar, the public terms do not support a reliable tasks-per-month estimate; observe how representative tasks change that meter before relying on it.
Check every window, not just the monthly total
A rolling window releases earlier usage as it ages out. A fixed window resets at an assigned time, which may follow a subscription anniversary or a calendar boundary. When several limits apply, the first exhausted window stops work even if another still has room.
Model choice and time of day can change consumption without changing the subscription price. Moonshot’s help-centre wording and developer documentation differ about weekly limits; this table follows the developer documentation for new and legacy plans. Recheck the official plan for your account before purchase.
The token and workload estimator can help build a task trace. Our Coding Plan has no monthly token allowance; requests may be slower at peak hours, and applicable terms are confirmed when access is granted. Request Coding Plan access to discuss your workflow and the price before paid use.
Sources and verification
Sources used for this resource. Verification dates describe our source checks, not provider effective dates.
Z.ai GLM Coding Plan documentation, Overview
Documents credit multipliers, five-hour and weekly allowances, reset rules and off-peak conditions for the reviewed Coding Plan.
Lists the reviewed monthly tier prices and subscription controls; its concurrency guidance gives tier ordering without numeric capacity.
Kimi Help Center, membership billing and plans
Describes legacy CNY memberships and a shared usage pool. Developer documentation supplies the distinction between legacy and new quota windows.
Kimi membership pricing, plans released in September
Lists the reviewed USD membership tiers and subscription controls; these prices do not normalize the shared allowance into tokens.
Kimi Code documentation, membership benefits
Documents the rolling five-hour window for new members, legacy weekly quotas and the shared monthly membership pool.
Kimi Code documentation, overview
Documents model eligibility and higher quota consumption for the longer-context model.
Lists tier prices, rolling five-hour and weekly windows, and a console usage bar without a numeric allowance.
Describes quota deduction at endpoint pricing and expiry of unused quota; it does not establish every reset or permitted workload.
Alibaba Cloud Model Studio, Coding plan overview
Documents model-call quotas, concurrent usage windows and reset rules. Its availability and throttling conditions belong to the dated review.
Alibaba Cloud Model Studio, Token Plan (Personal Edition)
Documents CNY prices, monthly Credits, concurrency guidance and the September 22 change from weekly to monthly quota.
Documents per-model dollar allowances, five-hour and weekly fractions, fallback behavior and workspace restrictions.
Lists the reviewed Go monthly price and subscription control; Go Plus pricing comes from the documentation.
DeepSeek API documentation, Models & Pricing
Documents per-token rates and peak/off-peak conditions. DeepSeek is included as a pay-as-you-go reference, not a coding subscription.
Claude Help Center, what is the Max plan?
Defines Max allowance relative to Pro, with session and weekly limits rather than a published absolute token quota.
Claude Help Center, what is the Pro plan?
Lists Pro pricing and usage conditions that vary with content, model and features.
Describes access on request, pricing, and usage with no 5-hour, weekly or monthly limit; requests may be slower at peak hours, subject to applicable terms; model availability may change.
Standard Thinking documentation, Coding Plan capacity and billing
Describes Coding Plan access on request and usage with no 5-hour, weekly or monthly limit; requests may be slower at peak hours, with configuration subject to applicable terms.
Lists plan pricing and model-specific API rates.
Continue exploring
TPMa rate with provider-specific rules
- Capacity
- PTU · MU · GSU
- Billing
- Units · GPU-hours
- Limits
- Windows · concurrency
Definitions before price comparisons
LLM capacity and plan terms defined
Definitions for inference capacity units, billing terms, workload estimates, and subscription limits, with provider-specific meanings kept explicit.
57.40public index, 8B SH
- 3B SH
- 51.53
- Retrieval & Classification
- 66.4
- Tools & Automation
- 84.0
Decision Index 0.3 public index, self-scored
Standard One SH scores every option in one pass
Open-weight decision models that return a probability for every option in one forward pass: 51.5 and 57.4 on the Decision Index 0.3 public index, self-scored with the official kit, with the detail and the limits.
3classification approaches
- Encoder
- Train on task labels
- LLM prompt
- Describe labels
- Decision model
- Score supplied options
Evaluate each on the same held-out workload
Fine-tuned classifier vs LLM: how to choose
Choose an encoder, prompted LLM or decision model by your data and labels; research and historical Standard One results use different measurement conditions.