Kimi Coding Plan: New Plans, Legacy Quotas and API Costs

If you searched for a Kimi K3 coding plan, the product to check is Kimi Code, supplied through Kimi membership. The current documentation says new plans are available: Plus and above include coding access, and new members no longer have a weekly rate window. They still have a five-hour window and monthly total quota. Legacy subscriptions keep different rules. Here is how to identify your plan, avoid calling the wrong billing route, and compare a subscription with metered access without treating old prices or small coding tests as current plan promises.

Updated October 4, 2026: replaced the July signup-status and tier-price snapshot with current new-versus-legacy rules. Removed the old Kimi-request/GLM-prompt conversion. Historical test costs below remain dated measurements; no new paid benchmark was run.

New plans are available; the July snapshot is not today's checkout

Kimi's membership documentation now says the new plans are available, with renamed tiers. Our earlier article described a July signup pause and a pending restructure. It should not be read as evidence that subscriptions are closed now.

The documentation names Go, Plus and Pro when explaining coding access: Go has no coding quota, Plus adds Kimi Code and K3 access, and Pro adds higher-tier capabilities. We could not independently verify a complete international checkout price table in this update. Use the official Kimi Code plan page for the currency, cycle and offer available to your account. The Chinese help center's RMB prices are not a substitute for a USD checkout.

No weekly limit does not mean unlimited coding

First establish whether your subscription is a new plan or a legacy one. The official membership page distinguishes them as follows:

Check monthly consumption in your subscription page, and coding rate-limit status in the Kimi Code Console. Multiple keys do not multiply your allowance. A new plan can still stop accepting coding work when the monthly total is exhausted.

Legacy renewals do not automatically mean migration to a new plan. The documentation says migration is optional. Before changing tiers, compare the displayed remaining term, quota rules and price rather than assuming a renamed plan is equivalent.

If your account offers Extra Usage, inspect its rate and spending-cap settings before enabling it. This is additional billed usage, not an extension of the included quota at no cost. We did not enable it or verify a charge; availability and regional terms belong to your account.

The model name and the subscription entitlement both matter

The model configuration page currently lists k3, k3-256k, kimi-for-coding and kimi-for-coding-highspeed. Importantly, kimi-for-coding now serves K2.8 Preview, not the K2.7 Code variant measured in our July benchmark.

k3-256k has a fixed 256K context. Full 1M K3 context and HighSpeed access require the appropriate higher tier. A correct-looking key can still fail when the requested capability exceeds your plan. Kimi's error reference advises reading the error text, since third-party clients may transform status codes.

When comparing results across dates, log the requested model and returned identity when available. An alias updated in place is not evidence that an old model became faster or cheaper. Start a fresh session when deliberately switching models so earlier cached context is not mistaken for the new model's behavior.

Kimi Code keys and metered API keys are different products

The service-endpoint documentation separates subscription access from Open Platform access. Obtain the key from the service you intend to use.

In a supported OpenAI-compatible coding client, set the coding base URL, use a Code Console key, and select an allowed ID such as kimi-for-coding. For Claude Code, use Kimi's tool-specific setup guide; do not paste a metered Open Platform configuration and assume membership covers it.

If authentication fails, check the key's product, protocol and model entitlement before retrying. A kimi-k3 ID from an Open Platform example is not automatically valid in a subscription client's model field.

What our historical coding tests can, and cannot, price

DataLLM Lab has recorded API coding tests using small Python tasks and hidden execution assertions. They are useful for a bounded task comparison, not for measuring Kimi Code subscription consumption. The route, token rates and run date matter.

Source: our dated measurement records and methodology. These are single-run, small-sample means and token-derived list-price costs, not provider invoices. The earlier July 16-17 K3 attempt completed eight tasks and could not evaluate the ninth after upstream rate limits; the July 30 record above is a later complete run. Neither is a test of today's K2.8 coding alias or a complete IDE workflow.

Use a verified subscription price and your own average metered cost:

break_even_tasks = monthly_subscription_price / average_cost_per_task

For illustration only, assume a $20 monthly plan; this is not a Kimi price quote. At the historical K3 rate of $6.33 per 1,000 small tasks, the arithmetic gives about 3,160 tasks/month. At an assumed $0.10 per agent task, it gives 200. The second cost is not measured, and neither calculation proves a plan permits that many tasks.

Use the total task bill divided by completed tasks for a spending comparison. A median individual call cost can understate a bill dominated by a few long turns. Agent tasks may include many model calls, so do not compare an API call count with an IDE turn count without measuring the relationship.

Kimi versus GLM: compare the same workload, not invented request counts

The previous article converted GLM prompts into estimated model calls and compared them with a Kimi request band. That conversion is removed: the current GLM Coding Plan calculates token-weighted credits, while Kimi separates new and legacy time-window rules and a monthly quota.

A practical comparison is to run the same allowed coding workflow, record completed tasks and interruptions, and read each product's usage meter before and after. Include code correctness, context size, cache behavior and actual monthly price. Our historical model table does not establish which subscription delivers more work per dollar.

Before paying or upgrading

1. Record the actual checkout currency, billing cycle and renewal price.
2. Identify new versus legacy quota rules and the capabilities your tier includes.
3. Verify that your tool uses the subscription endpoint and a Code Console key.
4. Measure representative complete tasks, including failed turns and tool calls.
5. Compare the resulting average cost with the monthly fee, then account for quota interruptions and any Extra Usage charges.

For metered K3 details, use the Kimi K3 API guide. For third-party access, DataLLM Lab's catalog is a separate billing route; it does not consume a Kimi membership entitlement.

FAQ