Skip to main content
RapidDev - Software Development Agency
AI API Limits & Performance Matrix12 min readVerified July 10, 2026

Kimi K2 API Rate Limits, Pricing & Performance (July 2026)

The legacy Kimi K2 family (kimi-k2-0905-preview and related slugs) reached end-of-life on May 25, 2026. The current successor is Kimi K2.6, Moonshot AI's flagship: 262K-token context, $0.95 input / $4.00 output per MTok with automatic caching at $0.16/MTok (~83% off). Tier 0 caps at 1.5M tokens/day; recharge $10 to unlock 200 RPM and unlimited daily tokens.

4.9Clutch rating
600+Happy partners
17+Countries served
190+Team members

Deprecated model

Sunsets May 25, 2026

Legacy Kimi K2 family (kimi-k2-0905-preview, kimi-k2-0711-preview, kimi-k2-turbo-preview, kimi-k2-thinking, kimi-k2-thinking-turbo) reached scheduled end-of-life on May 25, 2026. Legacy K2 pricing was $0.60 input / $2.50 output / $0.15 cached — no longer applicable. New projects must use Kimi K2.6 or K2.5. Old slugs may not resolve after the EOL date.

Migrate to:Kimi K2.6Current Moonshot AI flagship at $0.95/$4.00 with automatic caching at $0.16/MTok (83% off), 262K context, open weights (Modified MIT), and thinking mode — released April 2026.
Moonshot AIGenerally available

API model string

kimi-k2.6 (current flagship); kimi-k2.5 (cheaper option); kimi-k2.7-code (coding specialist, June 12, 2026). Legacy kimi-k2-* slugs: EOL May 25, 2026.

Context window

262K tokens (K2.6); 256K tokens (K2.7 Code)

Max output not published separately

Knowledge cutoff
not published
Released
Legacy K2 family: 2024–2025; K2.6: April 2026; K2.7 Code: June 12, 2026
Modalities
text in (K2.6 multimodal); text out; thinking modes available on K2.6 and K2.7

Last verified July 10, 2026

Rate limits by tier

Moonshot AI uses cumulative-recharge tiers for the Kimi API (OpenAI- and Anthropic-compatible at api.moonshot.ai/v1). Rate limits are per cumulative recharge tier. Legacy Kimi K2 reached EOL May 25, 2026; the limits below describe the current K2.6 / K2.5 / K2.7 family.

TierRequirementsRPMTPMRPDConcurrentNotes
Tier 0 (entry, no recharge)Account created; no recharge required to enter Tier 0not published1.5M tokens/day (hard daily cap at Tier 0)not publishednot publishedMust recharge at least $1 to fully activate account. Recharge $5 cumulative to receive a $5 voucher bonus. The 1.5M token/day cap is the main constraint at Tier 0 — blocks production workloads quickly.
Tier 1 ($10 cumulative recharge)At least $10 total cumulative recharge200 requests/minuteunlimited (daily cap lifted)unlimited50 concurrent requestsPractical entry tier for development and low-volume production. Daily token cap removed — 200 RPM and 50 concurrent are the active limits.
Tier 2 ($100 cumulative recharge)At least $100 total cumulative recharge5,000 requests/minuteunlimitedunlimited200 concurrent requestsProduction-ready tier. 5,000 RPM and 200 concurrent handle significant throughput for most SaaS use cases.
Enterprise / Tier 5+Contact sales@moonshot.cn for custom limits beyond Tier 5not publishednot publishednot publishednot publishedCustom rate limits, SLAs, and dedicated infrastructure available via Moonshot AI enterprise agreements.

Swipe the table sideways to see every limit column.

  • 1.OpenAI- and Anthropic-compatible API endpoint: api.moonshot.ai/v1.
  • 2.Rate limits are per cumulative recharge tier — tiers do not downgrade.
  • 3.Thinking tokens (K2.6 optional; K2.7 Code forced on) are billed as output tokens and inflate effective output volume — budget accordingly.
  • 4.Legacy K2 family (kimi-k2-0905-preview, kimi-k2-0711-preview, kimi-k2-turbo-preview, kimi-k2-thinking, kimi-k2-thinking-turbo) reached EOL May 25, 2026 — old slugs may not resolve after that date.

Limits verified against the Moonshot AI docs, July 10, 2026.

Token pricing

What you pay per million tokens (USD). Input and output are billed separately.

Input

$0.95

per 1M tokens

Output

$4.00

per 1M tokens

Cached input

$0.16

per 1M cached

Batch discount

40%

on batch jobs

  • Pricing shown is for Kimi K2.6 (current flagship). Legacy K2 pricing ($0.60 in / $2.50 out / $0.15 cached) is no longer applicable after EOL May 25, 2026.
  • Kimi K2.5 (cheaper option): $0.60 input / $3.00 output / $0.10 cached — approximately 37% cheaper across the board.
  • Kimi K2.7 Code: $0.95 input / $4.00 output (same as K2.6); forces thinking on — budget for higher output volume.
  • Automatic context caching at $0.16/MTok cache-hit (~83% off $0.95 standard) — no configuration required, Moonshot applies automatically.
  • Batch pricing: ~40% off — K2.6 batch approximately $0.57 input / $2.40 output per MTok.
  • Web search tool calls: $0.005/call plus result tokens billed as input on the following turn.
  • Thinking tokens are billed as output (no separate price) and can inflate volume significantly, especially K2.7 Code which forces thinking on.

Side-project chatbot

$8.75

per month

Assumptions

5M input tokens / 1M output tokens per month (K2.6 rates, no caching)

5 × $0.95 + 1 × $4.00 = $4.75 + $4.00 = $8.75/month

Mid-scale RAG pipeline

$71.82

per month

Assumptions

60M input / 12M output per month, 70% cached reads at $0.16/MTok (K2.6)

(18M × $0.95 + 42M × $0.16) / 1M + 12M × $4.00 / 1M = $17.10 + $6.72 + $48.00 = $71.82/month

High-throughput agentic system

$700

per month

Assumptions

400M input / 80M output per month (K2.6 rates, no caching)

400 × $0.95 + 80 × $4.00 = $380 + $320 = $700/month

Run your own numbers

Drag your real monthly token volumes and watch the bill update live — priced against rival models at the same usage.

30M
1M500M
5M
0.1M100M

Estimated kimi-k2.6 (current flagship); kimi-k2.5 (cheaper option); kimi-k2.7-code (coding specialist, June 12, 2026). Legacy kimi-k2-* slugs: EOL May 25, 2026. spend

$49/mo

Input: $29

Output: $20

30M in × $0.950 + 5M out × $4.00 = $49

Same volume, priced across models

  • kimi-k2.6 (current flagship); kimi-k2.5 (cheaper option); kimi-k2.7-code (coding specialist, June 12, 2026). Legacy kimi-k2-* slugs: EOL May 25, 2026.This model$49
  • Qwen3-Max$66
  • Grok 4$90
  • Mistral Large$90

Rivals priced at their published input/output rates for the same monthly volumes. Prompt caching is model-specific, so it is applied to kimi-k2.6 (current flagship); kimi-k2.5 (cheaper option); kimi-k2.7-code (coding specialist, June 12, 2026). Legacy kimi-k2-* slugs: EOL May 25, 2026. only. Estimates for comparison; real bills vary with request shape and long-context surcharges.

kimi-k2.6 (current flagship); kimi-k2.5 (cheaper option); kimi-k2.7-code (coding specialist, June 12, 2026). Legacy kimi-k2-* slugs: EOL May 25, 2026. vs the alternatives

Kimi K2.6 (successor) vs Grok 4.5, Qwen3-Max, and Mistral Large 3 — all live frontier-class models in the same capability tier.

Aspectkimi-k2.6 (current flagship); kimi-k2.5 (cheaper option); kimi-k2.7-code (coding specialist, June 12, 2026). Legacy kimi-k2-* slugs: EOL May 25, 2026.Grok 4Qwen3-MaxMistral Large
Context window262K tokens (K2.6)500K tokens (Grok 4.5)262K tokens (Qwen3-Max)128K tokens (Mistral Large 3)
Input $/MTok$0.95 (K2.6)$2.00 (Grok 4.5)$1.20 (Qwen3-Max)$2.00 (Mistral Large 3)
Output $/MTok$4.00 (K2.6)$6.00 (Grok 4.5)$6.00 (Qwen3-Max)$6.00 (Mistral Large 3)
Cache hit $/MTok$0.16 (~83% off, automatic)$0.50 (75% off) (Grok 4.5)not published (Qwen3-Max)available (Mistral Large 3)
Batch discount~40% offnone at launch (Grok 4.5)50% (Qwen3-Max)50% (Mistral Large 3)
Open weightsyes (Modified MIT)no (Grok 4.5)yes (Apache 2.0) (Qwen3-Max)partial (Mistral Large 3)
Free API tierno (recharge to activate)no (Grok 4.5)1M trial/model (Qwen3-Max)yes — experimental (Mistral Large 3)
Thinking modeoptional (K2.6) / forced (K2.7 Code)optional (Grok 4.5)optional (Qwen3-Max)Magistral only (Mistral Large 3)

Swipe the table sideways to see every model.

Hitting a 429? The playbook

The exact errors you'll see

HTTP 429 Too Many RequestsMoonshot API returns HTTP 429 on exceeding concurrent request limit, RPM cap, or daily token cap (Tier 0)OpenAI-SDK compatible error structure (same as standard OpenAI 429 response)

Why it happens & how to fix it

Tier 0 daily token cap (1.5M tokens/day) exhausted

Recharge at least $10 cumulative to unlock Tier 1 — this removes the daily token cap entirely and enables 200 RPM and 50 concurrent requests. The Tier 0 daily cap blocks production workloads quickly.

Concurrent request limit exceeded (50 at Tier 1 / $10 recharge)

Implement a request queue or semaphore to limit parallel in-flight requests. At $10 tier, cap concurrency at 40–45 to leave headroom. Recharge to $100 for 200 concurrent.

RPM exceeded (200 RPM at Tier 1, 5,000 RPM at Tier 2)

Add exponential backoff with jitter on 429 responses. Use automatic caching to reduce redundant calls for repeated-prefix workloads — cache-hit requests at $0.16/MTok still count toward RPM, but reduce total tokens and cost.

Thinking token volume spike (especially K2.7 Code with forced thinking)

Switch to K2.6 for non-coding tasks and use reasoning_effort=low for classification/routing. K2.7 Code forces thinking on — output volume can be 2–3× standard output volume.

Legacy K2 slug called after EOL (May 25, 2026)

Migrate to kimi-k2.6 — old slugs (kimi-k2-0905-preview, kimi-k2-0711-preview, kimi-k2-turbo-preview, kimi-k2-thinking, kimi-k2-thinking-turbo) may not resolve after the EOL date. Update the model parameter in your API calls.

Retry strategy

Honor HTTP 429 responses with exponential backoff: catch the error, wait 2^attempt seconds, then retry. The Moonshot API is OpenAI-SDK compatible — swap baseURL to https://api.moonshot.ai/v1 and update the api_key. Also compatible with @anthropic-ai/sdk. Add ±20% random jitter to prevent thundering-herd in multi-worker setups. Cap max retries at 5–6. For Tier 0 daily cap hits, no amount of retrying will help — the fix is to recharge to $10.

retry.ts
1import OpenAI from 'openai';
2
3const client = new OpenAI({
4 apiKey: process.env.MOONSHOT_API_KEY,
5 baseURL: 'https://api.moonshot.ai/v1',
6});
7
8async function chatWithRetry(
9 messages: OpenAI.Chat.ChatCompletionMessageParam[],
10 maxRetries = 6
11): Promise<string> {
12 for (let attempt = 0; attempt < maxRetries; attempt++) {
13 try {
14 const response = await client.chat.completions.create({
15 // Migrate from legacy kimi-k2-* slugs to current K2.6
16 model: 'kimi-k2.6',
17 messages,
18 });
19 return response.choices[0].message.content ?? '';
20 } catch (err) {
21 if (err instanceof OpenAI.RateLimitError && attempt < maxRetries - 1) {
22 const jitter = 1 + Math.random() * 0.2;
23 const wait = Math.pow(2, attempt) * 1000 * jitter;
24 console.warn(`Rate limit — waiting ${Math.round(wait)}ms (attempt ${attempt + 1})`);
25 await new Promise((r) => setTimeout(r, wait));
26 } else {
27 throw err;
28 }
29 }
30 }
31 throw new Error('Max retries exceeded');
32}

How to raise your limits

The ladder from the starter tier to enterprise — what each rung takes, and what it unlocks.

1

Tier 0 → activate account

Immediate on recharge.

Recharge at least $1 to fully activate. Account created without recharge has limited access; $5 cumulative spend earns a $5 voucher bonus.

Unlocks: Activated API access with 1.5M tokens/day cap.

2

Tier 1 — $10 cumulative recharge

Automatic on crossing the $10 threshold.

Reach $10 in total cumulative recharges on your Moonshot account (platform.moonshot.ai).

Unlocks: 200 RPM, 50 concurrent requests, unlimited daily tokens — daily cap removed.

3

Tier 2 — $100 cumulative recharge

Automatic on crossing the $100 threshold.

Reach $100 in total cumulative recharges.

Unlocks: 5,000 RPM, 200 concurrent requests — production-scale throughput.

4

Enterprise / Tier 5+

Custom negotiation — typically 1–3 weeks.

Contact sales@moonshot.cn with use case and volume requirements.

Unlocks: Custom RPM, concurrent, SLAs, and dedicated infrastructure beyond Tier 5 defaults.

Cut your token spend

Rely on automatic context caching

-83% on repeated-prefix input cost ($0.16 vs $0.95/MTok)

Moonshot AI applies automatic caching — no configuration needed. Cache-hits on stable prefix content (system prompts, retrieved documents, conversation history) are billed at $0.16/MTok. Highest impact in agentic loops with a large stable system prompt on every call.

Use K2.5 for cost-sensitive tasks

~37% cheaper across input and output vs K2.6

Kimi K2.5 at $0.60 input / $3.00 output / $0.10 cached is the budget option within the K2 family. Route simple tasks (classification, extraction, Q&A) to K2.5; escalate to K2.6 for nuanced generation or tool use.

Avoid K2.7 Code unless the task is coding-specific

Prevents 2–3× output token inflation from forced thinking

K2.7 Code forces thinking mode on for every request, significantly inflating output token volume (and therefore cost at $4.00/MTok). Use K2.6 with reasoning_effort=low for non-coding tasks.

Use batch API for async workloads

~40% off — K2.6 batch at approximately $0.57 input / $2.40 output per MTok

For classification, summarization, or analysis pipelines that tolerate latency, submit batch jobs. Batch cannot be combined with some other discounts — verify current batch rules in platform.moonshot.ai docs.

Minimize web search tool calls

Avoids $0.005/call charges accumulating unnoticed in agentic workflows

Each web search call costs $0.005 plus result tokens billed as input on the next turn. Aggregate multiple search intents into a single call per turn where possible.

Recharge to $10 early in development

Removes 1.5M token/day Tier 0 cap that blocks realistic testing

At $10 cumulative recharge, the daily token cap is removed entirely and you get 200 RPM and 50 concurrent — sufficient for most development and low-volume production workloads.

Self-host via open weights for high-volume or privacy-sensitive use

Potential cost reduction below $0.95/MTok at very high volume; full data control

K2.6 open weights are available on Hugging Face under a Modified MIT license. Third-party hosts (OpenRouter, Fireworks, DeepInfra) also offer K2 family at $1.15–$2.15/MTok blended per Artificial Analysis trackers — verify current rates before committing.

Frequently asked questions

Is Kimi K2 still available via API?

The legacy Kimi K2 family (kimi-k2-0905-preview, kimi-k2-0711-preview, kimi-k2-turbo-preview, kimi-k2-thinking, kimi-k2-thinking-turbo) reached end-of-life on May 25, 2026. Old slugs may no longer resolve. The current successor is Kimi K2.6 (launched April 2026) — update your model parameter to kimi-k2.6 and verify on platform.moonshot.ai.

What is the Kimi K2 API price?

Legacy Kimi K2 pricing ($0.60/$2.50/MTok) no longer applies after EOL. The current Kimi K2.6 flagship is $0.95 input / $4.00 output per MTok with automatic cache-hit pricing at $0.16/MTok (~83% off). K2.5, the budget option, is $0.60 input / $3.00 output / $0.10 cached.

How do I increase Kimi K2.6 rate limits?

Rate limits unlock automatically with cumulative recharges on your Moonshot account (platform.moonshot.ai): $10 cumulative gives 200 RPM and 50 concurrent requests with unlimited daily tokens; $100 cumulative gives 5,000 RPM and 200 concurrent. For custom limits beyond Tier 5, contact sales@moonshot.cn.

Is the Kimi K2 API free?

There is no free API tier. You must recharge your account to access the API. Recharging at least $1 fully activates your account; $5 cumulative earns a $5 voucher bonus. Tier 0 (pre-recharge) is capped at 1.5M tokens/day, which is insufficient for production use.

How does Kimi K2.6 compare to Qwen3-Max?

Both K2.6 and Qwen3-Max have 262K context windows. K2.6 is cheaper on output ($4.00/MTok vs Qwen3-Max's $6.00/MTok) and has stronger caching (automatic $0.16/MTok vs not published for Qwen). Qwen3-Max has a 50% batch discount vs K2.6's ~40%. Both have open weights. K2.6 is from a China-based provider — consider compliance requirements for EU/US enterprise.

What does HTTP 429 mean on the Moonshot/Kimi API?

HTTP 429 Too Many Requests means you exceeded your concurrent request limit, RPM cap, or (at Tier 0) the 1.5M tokens/day daily cap. If you are on Tier 0, no amount of backoff will fix a daily cap hit — you need to recharge to $10. For RPM/concurrent 429s, implement exponential backoff (wait = 2^attempt seconds with ±20% jitter) and retry. The Moonshot API is OpenAI-SDK compatible — swap baseURL to https://api.moonshot.ai/v1.

Can Kimi K2.6 be self-hosted?

Yes. Kimi K2.6 is open weights under a Modified MIT license, available on Hugging Face. Third-party inference hosts including OpenRouter, Fireworks, and DeepInfra also offer the K2 family at $1.15–$2.15/MTok blended (per third-party trackers — confirm current rates before committing). Self-hosting is viable for privacy-sensitive workloads or very high volume where managed API costs exceed infrastructure costs.

Can RapidDev help optimize Kimi K2 costs or migrate from legacy K2?

Yes. RapidDev handles caching architecture, gateway design, and model migration for production AI APIs. If you need help migrating from legacy K2 slugs to K2.6, reducing spend via automatic caching, or managing rate limits in a multi-model pipeline, book a free scoping call at rapidevelopers.com/contact.

RapidDev

We build AI apps that don't hit rate limits

  • Retry, backoff & caching built in
  • Multi-provider fallback routing
  • Fixed price, you own the code
Get a free estimate

30-min call. No commitment.

Still weighing your options?

Talk to a team that ships on all of these platforms. A free consultation gets you an honest recommendation for your specific project — even if the answer is a tool, not us.

Book a free consultation

We put the rapid in RapidDev

Need a dedicated strategic tech and growth partner? Discover what RapidDev can do for your business! Book a call with our team to schedule a free, no-obligation consultation. We'll discuss your project and provide a custom quote at no cost.