Batch API Inference now available for K3 models at 60% discount pricing
Developers can now use the Batch API for non-real-time K3 model tasks at 40% cost savings (60% of standard price), enabling more economical processing of large-scale workloads.
Evidence
+Batch API Inference: kimi-k3.1-pro and kimi-k3 support batch inference for large-scale tasks that do not require real-time responses, billed at 60% of the standard price.
Default model in examples changed from kimi-k3 to kimi-k3.1-pro
All code examples now use the new K3.1 Pro model by default, requiring developers to explicitly specify alternative models if they want to use older versions.
Context caching terminology updated from 'Cache Write' to automatic billing model
Documentation clarifies that context caching is automatically enabled for repeated request prefixes with default 5m TTL, simplifying the billing model explanation.
Evidence
−Note: Cache Write refers to the cost of writing request prefixes into the context cache. The Kimi API automatically caches repeated request prefixes, and cache entries come in two time-to-live (TTL) tiers — 5min and 1h. If no TTL is specified, the 5min tier applies by default.
+Note: The Kimi API automatically enables context caching billing for repeated request prefixes, with a default TTL (time-to-live) of 5m.
K2 series models being deprecated with phased retirement plan
K2 series will be unavailable to new users starting October 31, 2026, and fully retired by December 31, 2026. Existing users must migrate to newer models.
Evidence
+starting October 31, 2026, K2 series models will no longer be available to newly registered users; existing users are not affected. We plan to retire all K2 series models by the end of December 2026, please prepare your migration in advance.
API documentation clarified with specific endpoint naming and branding changes
Developers integrating with Kimi API will see official naming of the OpenAI-compatible and Anthropic-compatible endpoints, and updated product branding references.
Evidence
−Kimi API lets you interact with Kimi models and is compatible with both the OpenAI and Anthropic API formats.
+Kimi API Platform provides access to Kimi models through the OpenAI-compatible Responses API and the Anthropic-compatible Messages API.
Kimi Code and Codex added to list of Kimi K3 use cases alongside Claude Code
Users are now informed that Kimi K3 can be used for Kimi Code and Codex programming agent scenarios, establishing Kimi's own agent solutions as primary use cases.
Evidence
−It is well suited for programming agent scenarios such as Claude Code, as well as knowledge work and deep reasoning.
+It is well suited for programming agent scenarios such as Kimi Code, Codex, and Claude Code, as well as knowledge work and deep reasoning.
Repositioned existing content under K2 Series Models category
Clarifies that the original billing model (input/output usage-based) applies specifically to K2 Series, distinguishing it from the new K3 Series caching model.
Evidence
−Model inference APIs bill both Input and Output based on usage.
+K2 Series Models
Model inference APIs bill both Input and Output based on usage.
Introduced K3 Series Models with context cache pricing structure
Users can now use K3 Series models with automatic request prefix caching (5min and 1h TTL options), enabling cost savings through cached input billing at lower rates than regular input.
Evidence
+K3 Series Models
Note: Cache Write refers to the cost of writing request prefixes into the context cache. The Kimi API automatically caches repeated request prefixes, and cache entries come in two time-to-live (TTL) tiers — 5min and 1h. If no TTL is specified, the 5min tier applies by default. Input that hits the cache within its lifetime is billed only at the Cached Input Price: each hit automatically refreshes the cache entry's lifetime, with no additional cache write charge.
Cache writes for K3 models billed separately per TTL tier with no charge on cache hits
K3 model users benefit from lower costs on repeated requests through cache hits, with cache write costs transparent and cache hits charged only at reduced cached input rates.
Evidence
−Model inference APIs bill both Input and Output based on usage.
+For K3 series models, cache writes are billed separately per TTL tier (5min / 1h); cached input is billed at the cache-hit price only, with no additional cache write charge.
Billing method changed to 50/50 split between cash and vouchers starting September 14
All users without signed contracts will have account usage deducted at 50% from cash and 50% from vouchers instead of previous method. When either balance depletes, remaining usage draws from the remaining balance type.
Evidence
+Starting at 00:00 on September 14, account usage on the Kimi API Open Platform (platform.kimi.ai) will be deducted at a uniform ratio of 50% from cash and 50% from vouchers. When either the cash or voucher balance is depleted, all subsequent account usage will be deducted from the remaining balance type.