Kimi K2.6

Marathon coding agents with visual context

Chat
Multi-Modal
Thinking
JSON Mode
Tool Calling

Pricing

Run instantly. Pay only for what you use.

Input
$0.6/ M
Output
$3.2/ M
Cache read
$0.10/ M

About Kimi K2.6

Kimi K2.6 is Moonshot AI's open-weight agentic flagship under a Modified MIT license: a 1T-parameter MoE with 32B active per token across 384 experts, a 262K-token context window, and native vision through its 400M-parameter MoonViT encoder, with thinking and instant modes.

It is built for marathon autonomy: runs sustain 4,000+ coordinated tool calls over 12+ hours, and its Agent Swarm architecture scales to 300 parallel sub-agents — the model keeps a plan where others lose the thread.

It posts 80.2% on SWE-bench Verified, 58.6 on SWE-bench Pro, and 66.7 on Terminal-Bench 2.0, and scores 44 on the Artificial Analysis Intelligence Index — the highest of any model in the Entrim catalog.

1Ttotal parameters, 32B active

A 384-expert MoE — frontier scale billed at 32B-active compute.

12+hours of autonomous running

Sustains 4,000+ coordinated tool calls in a single run without losing the plan.

300parallel sub-agents

Agent Swarm coordination can produce 100+ files from a single prompt.

Key capabilities

Kimi K2.6 is strongest where agents run longest — overnight refactors, multi-agent pipelines, and UI work that needs a model to hold a plan for hours, not minutes.

Marathon agentic coding

80.2% on SWE-bench Verified and 66.7% on Terminal-Bench 2.0 — plans hold across thousands of steps.

Agent Swarm orchestration

300 sub-agents and 4,000 steps per run — parallel decomposition is part of the model, not your framework.

Native vision

The 400M-parameter MoonViT encoder reads screenshots and design files — front-end work grounded in real UI.

Thinking and instant modes

Deep reasoning for the hard steps, instant replies for quick ones — switchable per request.

Where this model fits

Best for
  • Marathon coding agents

    Sustains 4,000+ tool calls across 12+ hour autonomous runs without losing the plan.

  • Multi-agent orchestration

    Agent Swarm scales to 300 parallel sub-agents and 4,000 steps — one prompt can produce 100+ files.

  • Frontier-adjacent engineering

    80.2% on SWE-bench Verified — open-weight results within a point of closed frontier models.

  • Front-end work from screenshots

    MoonViT vision grounds UI tasks in real screenshots and design images — visual context stays in the loop.

Avoid for
  • Cost-sensitive, high-volume fleets

    Premium 1T-class pricing. For always-on loops where per-token cost dominates, use Qwen 3.6 35B-A3B.

  • Latency-critical chat

    Throughput is modest for its class and thinking traces run long — about 1.7× the median output tokens in Artificial Analysis testing. Use GPT OSS 120B for fast replies.

  • Audio input or speech understanding

    Inputs are text and image. Transcribe audio upstream before sending it to the model.

Performance and
benchmarks

Entrim performance for Kimi K2.6 under production load in the past few days.

Throughput (tokens / sec)

avg. 48.6tokens/s

Time to first token (TTFT)

avg. 980ms

Public model benchmarks

80.2%SWE-bench Verified

Resolves real GitHub issues end-to-end in agentic coding runs.

58.6SWE-bench Pro

Harder, contamination-resistant software engineering tasks.

66.7%Terminal-Bench 2.0

Autonomous work in a real terminal — builds, debugging, ops tasks.

Run Kimi K2.6 in minutes

Test this model with $25 free credit. Use Entrim's OpenAI-compatible API to call Kimi K2.6, compare output quality, latency, and request cost, then decide if it fits your workload.

Code examples

curl -X POST "https://api.entrim.ai/v1/chat/completions" \  -H "Content-Type: application/json" \  -H "Authorization: Bearer $ENTRIM_API_KEY" \  -d '{    "model": "moonshotai/Kimi-K2.6",    "messages": [      {        "role": "user",        "content": "Why do they call it a building if it is already built?"      }    ]  }'

SDKs & docs

Use familiar SDK patterns with quickstart examples for common production setups.

OpenAI-compatible

Keep the request format your team already knows.

No cold starts

Production requests are served without model spin-up delays.

Not sure this is the right model?

Explore related models that may be a better fit for your workload.

DeepSeek V4 Flash

Version 0731

High-throughput agentic work with 1M context

1M
FP4
Thinking
Tool Calling
Input / M
$0.09
Output / M
$0.17
Cached / M
$0.015

Tool-heavy coding at 3B-active cost

262K
FP8
Vision
Multimodal
Video
Input / M
$0.05
Output / M
$0.25
Cached / M
$0.025

Long-horizon engineering with planning depth

200K
FP8
Thinking
Input / M
$0.9
Output / M
$3.0
Cached / M
$0.15

Run Kimi K2.6 with the Most Cost-Effective LLM Inference

Use $25 free credits to test prompts, token sizes, latency, throughput, output quality, and request cost before moving traffic.
All services are online

© 2026. Entrim. All Rights Reserved.

Privacy policyTerms of service