KWAIPILOT KAT-CODER256K CONTEXTAGENTIC CODINGCNY SOURCE
KAT-Coder-Pro V2.5 API Pricing
KAT-Coder-Pro V2.5 is Kwaipilot's flagship agentic coding model, served on Kuaishou's StreamLake API. It costs CNY 5/M input and CNY 20/M output ($0.70 and $2.82 at the project rate of 7.10), with cache reads at CNY 1/M and cache writes free. Note the direction of travel: this is 2.4x more expensive per token than the Pro V2 it follows.
Input - per 1M tokens
$0.70/M
Was CNY 2.1 on V2 +138%
Output - per 1M tokens
$2.82/M
Was CNY 8.4 on V2 +138%
Cached input - per 1M tokens
$0.14/M
Cache writes free -80%
Effective - agentic blend
$0.45/M
92/8 split - 82% cache
§ 01 / TERMINAL
Run the numbers.
Live calculator pre-loaded with current KAT-Coder-Pro V2.5 rates. Tweak the workload split, then share the URL to share the calculation.
$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
Real-world presets.
CODING AGENT
Codebase-scale migration
$0.057/task
LONG DOC
Reading 100-page contracts
$0.066/doc
RAG SUPPORT
Ticket triage
$0.00475/ticket
ASSISTANT
Research planning turn
$0.00999/turn
§ 03 / TOKENIZER
Paste text. See tokens. See cost.
Estimate · kwaipilot-tokenizer-estimate · ≈3.85 chars/token Auto-counts as you type
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
Characters 589
Words 96
Tokens (estimated) 153 tokens
Cost as input · uncached $0.00011 USD
Cost as output · uncached $0.00043 USD
Cost as cached input $0.00002 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| KAT-Coder-Pro V2.5 Current | $0.70 cache $0.14 | $2.82 | $0.45 agentic 92/8 | 256K | Repo-scale agentic coding tasks |
| KAT-Coder-Air V2.5 | $0.14 cache $0.03 | $0.56 | $0.09 same family, fast tier | 256K | High-volume agent loops |
| KAT-Coder-Pro V2 | $0.30 cache $0.06 | $1.18 | $0.19 previous Pro, still listed | not listed | Same Pro line at the old rate |
| LongCat-2.0 | $0.70 cache $0.01 | $2.82 | $0.35 identical list rates, deeper cache | 1M | Agentic coding with 1M context |
| Grok Build 0.1 | $1.00 cache $0.20 | $2.00 | $0.48 coding specialist | 256K | Coding agents with cheap output |
| Kimi K2.7 Code | $0.95 cache $0.19 | $4.00 | $0.62 Chinese coding peer | 262K | Repo-scale Chinese coding workloads |
Frequently asked.
Practical KAT-Coder-Pro V2.5 pricing questions, including why this generation costs more than the last.
Q · 01 What does KAT-Coder-Pro V2.5 cost? +
The StreamLake price table lists
CNY 5 per 1M input tokens, CNY 20 per 1M output tokens and CNY 1 per 1M cached input tokens, with cache writes free. At the project rate of 7.10 CNY/USD that is $0.704225, $2.816901 and $0.140845.Q · 02 Why is V2.5 more expensive than V2? +
Kwaipilot does not explain the increase, but the table is unambiguous: Pro V1 and Pro V2 are both
CNY 2.1/8.4, while V2.5 is CNY 5/20 - a 2.4x rise per token. Both older rows are still listed and callable, so if V2.5's capability gain does not pay for itself on your workload, Pro V2 remains available at the old rate.Q · 03 How capable is it? +
Kwaipilot's product page reports SWE-bench-Pro 65.2% for Pro V2.5 against 42.4% for Air V2.5. That is the vendor's own published number, not an independent run - we have not verified it against a neutral harness.
Q · 04 What are the context and output limits? +
The product page lists a
256K context window and 80K maximum output for both V2.5 models, with streaming, context caching, MCP and function calling supported across 20+ programming languages.Q · 05 Are cache writes really free? +
Yes - the product page marks 缓存写入 (cache write) as 免费 (free) for both V2.5 models, and the price table leaves the column blank. Only cache reads are billed, at
CNY 1/M, an 80% discount versus a miss. A miss still bills at the normal input rate, which is what our effective blend assumes.Q · 06 Where does the API live? +
On StreamLake (万擎 / Vanchin), Kuaishou's own cloud platform - the same console that sells third-party models. For KAT-Coder this is the first-party channel, so these are Kwaipilot's own rates rather than a reseller markup.
Q · 07 How does it compare to other coding models? +
At an effective
$0.45/M it sits within 6% of Grok Build 0.1 and 38% under Kimi K2.7 Code. LongCat-2.0 carries identical CNY 5/20 list rates but a far deeper cache discount, which puts it 21% cheaper on the same blend - and gives 1M context against this model's 256K.Q · 08 Is there a subscription instead of per-token billing? +
Kwaipilot sells a KwaiKAT Coding Plan billed per prompt rather than per token. That is a platform subscription, not a model rate, so it is not recorded here - every figure on this page is the pay-as-you-go token price.