Short answer: For a new integration using the providers’ own APIs, Moonshot says Kimi K2.5 was retired across its platform on August 31, 2026, while Z.ai’s pricing page lists GLM-5. In that Moonshot-versus-Z.ai first-party API comparison, GLM-5 is the currently listed option of the two—not a proven performance winner. This retirement notice does not establish whether Kimi K2.5 is available through other serving providers. A third-party routing page currently lists K2.5 with several provider options, but availability and price depend on the specific route and account; verify those terms directly. (Moonshot API documentation; Z.ai pricing; OpenRouter model listing)
At a glance
| Decision point | Kimi K2.5 | GLM-5 | What it means for a production choice |
|---|---|---|---|
| First-party API status | Moonshot documents K2.5 as retired across its platform on August 31, 2026 | Listed on Z.ai’s pricing page |
For a new integration using these first-party routes, confirm GLM-5 access for your account. Moonshot’s retirement does not establish K2.5 availability on other routes. (Moonshot API documentation; Z.ai pricing) |
| Input types | Moonshot’s launch announcement described native vision and visual reasoning | Z.ai documents text input and text output for GLM-5 |
The launch announcement describes K2.5’s intended capabilities, not current Moonshot API availability. (Moonshot announcement; GLM-5 documentation) |
| Context and output | A third-party comparison reports a 256K context window; this is not a current Moonshot API promise | Z.ai lists 200K context and 128K maximum output |
Use provider-documented limits for the exact model and route you plan to call. (GLM-5 documentation) |
| API price | No current Moonshot K2.5 price comparison applies after its first-party retirement. Other routes may list different prices; verify the exact provider and terms | Z.ai lists $1 per 1M input tokens, $3.20 per 1M output tokens, and $0.20 per 1M cached input tokens. Its rate card separately marks cached-input storage as “Limited-time Free”; check the current card for applicable terms. |
These are not matched current prices across comparable routes. Do not treat third-party K2.5 listings as Moonshot’s rate card. (Z.ai pricing; OpenRouter model listing) |
| Performance verdict | No current, version-matched comparison establishes a winner | No current, version-matched comparison establishes a winner | Choose a supported route and test the exact model versions on your workload rather than relying on a universal ranking. |
What the available evidence says about each model
Kimi K2.5: historical capabilities and route-specific availability
When Moonshot announced Kimi K2.5’s API availability on February 4, 2026, it described the model as supporting native vision, visual reasoning, tool calling, and long-context reasoning. Moonshot’s later platform documentation says Kimi K2.5 was retired across its platform on August 31, 2026. The retirement notice applies to Moonshot’s platform; it does not, by itself, establish availability or pricing through other serving providers. (Moonshot announcement; Moonshot API documentation)
A third-party comparison also positioned Kimi K2.5 for image and video work and parallel-agent workflows. It reported a 256K context window and described an “Agent Swarm” feature. These are competitor-article claims, not current API guarantees or results from a controlled comparison; that article does not provide primary documentation for the agent feature or production API instructions.
A third-party routing listing currently names Kimi K2.5 and multiple providers. That listing is evidence of what the routing page presents, not a guarantee that a particular provider will serve your account, meet your requirements, or offer a specific price. Verify the exact route, model ID, price, and terms before relying on it. (OpenRouter model listing)
Practical implication: If your use case depends on Kimi’s visual or agent capabilities, check whether a provider currently offers the exact K2.5 model on a route that meets your requirements. Do not transfer K2.5’s historical claims, limits, or prices to a newer Kimi model.
GLM-5: documented text, coding, and agent capabilities
Z.ai describes GLM-5 as a model for agentic engineering and long-range agent tasks. Its documentation lists text input and output, a 200K context length, a maximum output of 128K tokens, multiple thinking modes, streaming, function calling, context caching, and structured output. These documented capabilities do not prove GLM-5 will perform better than Kimi K2.5 on a particular task. (GLM-5 documentation)
Z.ai’s pricing page lists GLM-5 at $1 per million input tokens and $3.20 per million output tokens, with cached input at $0.20 per million tokens. The same rate-card row lists cached-input storage as “Limited-time Free.” The pricing page does not specify the promotion’s eligibility or end date in the retrieved row, so check the current rate card for applicable storage terms. Prices are listed in U.S. dollars. The page also lists newer models, including GLM-5.3, so GLM-5 is not the only GLM option to evaluate for a new deployment. (Z.ai pricing)
Which should you choose?
- For a new integration limited to Moonshot’s and
Z.ai’s first-party APIs: GLM-5 is the currently listed option of this exact pair in the evidence cited here: Moonshot documents K2.5 as retired, andZ.ailists GLM-5. Confirm GLM-5 availability, limits, and pricing for your account and intended route. (Moonshot API documentation;Z.aipricing) - If you are considering a third-party Kimi K2.5 route: Check the provider’s current model ID, callability, pricing, and terms. Moonshot’s first-party retirement notice does not answer those questions for other routes. (OpenRouter model listing)
- For image or video input:
Z.aidocuments GLM-5 as text-in/text-out. Moonshot announced native vision for Kimi K2.5, but retired it from its own platform. For another K2.5 route—or a currently supported replacement—verify the provider’s actual input format and availability. (GLM-5 documentation; Moonshot API documentation) - For coding or long multi-step text tasks: GLM-5’s documented function calling, thinking modes, and long context make it a candidate to test. Those capabilities do not establish a benchmark win over Kimi K2.5. (GLM-5 documentation)
- For a historical comparison or migration decision: Treat Kimi K2.5 claims and prices from older comparison articles as source- and route-specific, not as current Moonshot API terms. Third-party route prices, if available, need to be checked separately against the specific provider and GLM-5 route you would use.
GLM-5 API cost example
Using Z.ai’s listed rates, a hypothetical request with 100,000 uncached input tokens and 20,000 output tokens would cost:
- Input: 100,000 ÷ 1,000,000 × $1.00 = $0.10
- Output: 20,000 ÷ 1,000,000 × $3.20 = $0.064
- Estimated uncached total: $0.164, before any other charges.
This is an uncached estimate calculated from the listed per-token rates, not a measured workload result. It assumes all input tokens are billed at the standard input rate, not the cached-input rate. Z.ai lists cached input separately and marks cached-input storage “Limited-time Free”; check the current rate card for applicable terms. (Z.ai pricing)
How to compare a supported Kimi route with GLM-5
The sources cited here do not establish a current, controlled Kimi K2.5-versus-GLM-5 performance result for matched API routes. If you’re choosing between GLM-5 and a Kimi route that currently supports the exact model you need, run a small evaluation before committing:
- Choose representative tasks. Use real examples from your application, such as code changes, multi-step tool use, or visual analysis if that is part of the workload.
- Keep the setup comparable. Use the same task requirements, input data, tool definitions, and success criteria. Record the model IDs and API routes so results don’t get mixed across versions or providers.
- Score the outcome. Record task completion, correctness, required human fixes, and important failure modes. Measure latency if it affects your users.
- Calculate cost to an acceptable result. Include input and output tokens across retries or follow-up turns, rather than comparing input rates alone. Use the rates for the routes actually tested.
- Verify production constraints. Check current model availability, rate limits, data-handling terms, and provider-specific behavior in the documentation for each chosen route.
This is a proposed evaluation—not a test performed for this article. The available research does not include a matched production API test of Kimi K2.5 and GLM-5.
Bottom line
The availability verdict is specific to Moonshot’s first-party Kimi API versus Z.ai’s GLM-5 API. Moonshot documents Kimi K2.5 as retired across its platform from August 31, 2026, while Z.ai’s pricing page lists GLM-5. That does not establish that K2.5 is retired on every serving route: third-party provider listings should be checked for actual account-level availability, pricing, and terms. For a new project, choose models currently supported on your intended routes and compare those exact versions on your workload. GLM-5 has documented text, function-calling, and long-context features, but the evidence here does not establish a universal performance winner. (Moonshot API documentation; Z.ai pricing; OpenRouter model listing)
Sources
Pricing — Overview — Z.AI Developer Documentation
Use Kimi in OpenClaw — Kimi API Platform
Model List — Kimi API Platform
Kimi K2.5 API is now available — Kimi K2 — Kimi Forum
GLM-5 Overview — Z.AI Developer Documentation