Kimi K3: Moonshot's 2.8T Frontier Model — API, Pricing, and 1M Context Guide
Kimi K3 is Moonshot's most capable model — 2.8T parameters, 1M-token context, native vision, and reasoning_effort thinking mode. Official API pricing, Python examples, and limits from platform.kimi.ai docs.
Update — July 16, 2026: Moonshot AI launched Kimi K3 — and platform.kimi.ai now documents the official specs: 2.8 trillion parameters, 1M-token context, native vision, and a kimi-k3 API model ID at https://api.moonshot.ai/v1. Launch-night press placed early benchmarks above Claude Opus 4.8 and below Fable 5; this guide focuses on what Moonshot's platform docs confirm for builders — pricing, API patterns, and production limits — not leak-thread speculation.
TL;DR — official specs and quick answers
Question
Answer (official platform docs, Jul 16, 2026)
What model ID?
kimi-k3
How big is it?
2.8T parameters — Kimi's most capable model to date
Context window?
1,048,576 tokens (1M) — flat pricing, no length tiers
Architecture?
Kimi Delta Attention (hybrid linear attention) + Attention Residuals
Vision?
Native visual understanding — base64 or ms://<file-id> only; no public URLs
Being updated — not recommended for production near term
What Moonshot shipped — and what the platform docs confirm
K3 is Moonshot's most capable model to date, positioned for frontier workloads: software engineering, knowledge work, and deep reasoning. The pre-launch leak window (July 14–16) correctly flagged scale and architecture direction; official platform documentation now pins the numbers:
2.8 trillion parameters (not the "~2.5T" single-source leak figure)
Kimi Delta Attention — hybrid linear attention — plus Attention Residuals
1M-token context (1,048,576 tokens exactly)
Native visual understanding for multimodal agent loops
Launch-night reporting from Financial Times and X still describes K3 as open-weight, rolling out on kimi.com and the Kimi app for Mainland China users first — continuing Moonshot's pattern from Kimi K2.7 Code. For API access today, the authoritative source is platform.kimi.ai — keys, playground, and docs — not third-party aggregators.
Flat pay-as-you-go — no tiering by context length. Context caching is automatic for regular requests.
Token type
Price (per 1M tokens)
Input — cache hit
$0.30
Input — cache miss
$3.00
Output
$15.00
Cache strategy: With a 1M-token window, keep long system prompts, tool definitions, and document prefixes unchanged across requests. Moonshot's automatic caching rewards stable prefixes — the same pattern that made DeepSeek V4 Pro's permanent API discount interesting for agent loops, but at frontier reasoning depth.
Compare against K2.7 Code's published rates ($0.95/M input cache miss, $4/M output) in our K2.7 guide when choosing a Kimi model for cost-sensitive coding vs K3 frontier reasoning.
K3 always runs in thinking mode. Use reasoning_effort — not the K2.x thinking parameter.
python
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["MOONSHOT_API_KEY"],
base_url="https://api.moonshot.ai/v1",
)
response = client.chat.completions.create(
model="kimi-k3",
reasoning_effort="max", # only supported level today; default if omitted
messages=[
{"role": "system", "content": "You are a senior software engineer."},
{"role": "user", "content": "Refactor this handler to use async/await."},
],
max_completion_tokens=8192,
)
message = response.choices[0].message
# Parse final answer from content — not reasoning_contentprint(message.content)
Streaming reasoning and content separately
Streaming emits reasoning_content deltas (chain-of-thought) and final content deltas in separate streams:
Set strict: true on the schema. Do not parse reasoning_content for the JSON payload.
Vision — base64 only (no public URLs)
Vision content must be an array of objects. Use local base64 encoding or upload files and reference via ms://<file-id>. Public image URLs are not supported.
For video, use ms://<file-id> after uploading through Moonshot's file API — same constraint as MCP tool-use workflows where agents pipe local artifacts rather than hotlinked URLs.
Key API capabilities builders should know
Thinking mode (reasoning_effort)
K2.x pattern
K3 pattern
thinking: { "type": "enabled" }
reasoning_effort: "max" (top-level field)
Optional disable
Always enabled — cannot turn off
—
Only max supported today; more levels coming
Partial mode — continue from a prefix
Send an assistant message with partial=True to continue generation from a prefix — useful for autocomplete-style agent harnesses and loop engineering patterns where you steer mid-generation.
Tool choice and multi-turn
Set tool_choice="required" (or other OpenAI-compatible values) when tools must fire
In multi-turn tool loops, return the complete assistant message unchanged — do not strip tool calls or reasoning fields Moonshot expects round-tripped
Dynamic tool loading
Put tool definitions in a system message without a content field — Moonshot supports loading tools dynamically without bloating the visible system prompt.
Official tools via Formula API
Moonshot exposes first-party tools through the Formula API (/tools, /fibers). Web search is being updated — platform docs do not recommend it for production near term. Prefer your own retrieval or MCP servers until the search tool stabilizes.
Fixed parameters and limits — do not send these
Moonshot fixes several sampling parameters on K3. Omit them from requests — the server applies defaults:
Parameter
Fixed value
temperature
1.0
top_p
0.95
n
1
presence_penalty
0
frequency_penalty
0
Limit
Value
reasoning_effort
max only (for now)
max_completion_tokens default
131,072
max_completion_tokens maximum
1,048,576
Vision URLs
No public URLs — base64 or ms:// only
Multi-turn
Return full assistant message unchanged
Web search (Formula)
Not production-ready — being updated
How the leak window mapped to official specs
Before platform docs went live, K3 spent July 14–16 leaking across Moonshot's billing pages and X — and several predictions held:
Leaked (Jul 14–16)
Official / launch (Jul 16)
"~2.5T MoE parameters"
2.8T per platform docs
"New architecture, not scaled K2.6"
Kimi Delta Attention + Attention Residuals
"~1M token context"
1,048,576 tokens confirmed
"Community hopes for open-source Kimi 3.0"
Launch reporting: open-weight release
"~Fable tier, slightly worse" (tester chatter)
Press band: above Opus 4.8, below Fable 5
The recharge campaign page that briefly appeared July 14 on Moonshot's billing domain — pointing to a July 15 China-time start — was the strongest pre-launch artifact. The launch landed evening July 16, a short slip from the leaked date.
Treat benchmark placement as directional until Moonshot publishes an official card and independent evaluators run Senior SWE-bench-class suites. Launch-night X threads are not vendor benchmark cards.
K3's $15/M output pricing reflects frontier reasoning depth — budget agent loops that fire hundreds of tool calls may still favor K2.7 or local open models via OpenCode until you measure quality delta on your workload.
Safety and agent deployment
K3's launch lands the same week as Anthropic's agentic misalignment report, which tested Kimi K2.6 in simulated fraud scenarios. A larger model with 1M context, always-on reasoning, and tool APIs raises the same permission and audit questions — especially if open weights follow the K2 release pattern. The deployer owns safety configuration, not the vendor. Same lesson as Claude Memory Heist: long-context agents with web fetch need explicit guardrails.
Why "beats Opus 4.8" still matters for open-weight
If independent benchmarks confirm launch-night placement, K3 becomes evidence that open-weight frontier models can compete with closed APIs at the top of the scale — not just in the budget tier where DeepSeek and earlier Kimi releases competed. It ships alongside Inkling's 975B open MoE and Gemma 4 updates — but 2.8T is a different magnitude than any of them.
Summary
Kimi K3 launched July 16, 2026 as Moonshot's most capable model: 2.8T parameters, Kimi Delta Attention + Attention Residuals, 1M-token context, and native vision — accessible via kimi-k3 at https://api.moonshot.ai/v1. Official pricing is $0.30/$3.00/M input (cache hit/miss) and $15/M output, flat across context length with automatic caching. Builders should use reasoning_effort: "max" (not K2.x thinking), stream reasoning_content separately from content, and respect fixed sampling params and vision URL constraints. Launch press placed K3 above Opus 4.8 and below Fable 5; platform docs are the source of truth for API integration — benchmark cards and global weight downloads are still the next milestones to watch.
API specs, pricing, and limits accurate as of July 16, 2026, per Moonshot Kimi platform documentation. Verify current rates and model availability on platform.kimi.ai before planning production workloads — benchmark cards, web search tooling, and additional reasoning_effort levels were still rolling out at publication time.
