Skip to contentT/TTokens or Towers

Compare

LeaderboardEvery tracked model, ranked and priced side by sideBenchmarksWhat each test measures and who leads itPricesAPI, media, and hardware rates with a cost calculator

Plan

Stack plannerSix questions to a priced local and hosted stackShortlistOur picks for subscriptions, APIs, and hardwareEnterpriseSeat plans against self-hosting for a whole team

Build

HardwareGPUs and workstations, and which models they fitHostsWhere to rent open-weight inference, and for how muchHarnessesCoding agents and the models they drive

Lab

FindingsResults from our own evaluationsScanAudit a repo’s AI usage in the browserMethodWhere the numbers come from and how we check them
Data checked September 25, 2026
Data checked September 25, 2026
Plan a stack
T/TTokens or Towers

An independent guide to AI models, what they cost, and the hardware to run them. No account, no API key, no sponsors.

Compare

  • Leaderboard
  • Benchmarks
  • Prices

Plan

  • Stack planner
  • Shortlist
  • Enterprise

Build

  • Hardware
  • Hosts
  • Harnesses

Lab

  • Findings
  • Scan
  • Method
Data checked September 25, 2026 · Every number links to its sourceRSS · Source on GitHub
Leaderboard/Kuaishou
Kuaishou

Kuaishou · KAT-Coder V2.5

KAT-Coder-V2.5-Dev

Open weightsGenerally availableApache-2.0

Small open-weight (Apache-2.0) 35B/3B-active MoE coding model from Kwaipilot for local agentic coding.

AA Intelligence
—
Price
Self-host
Output speed
—
Context window
262K tokens

Coding

  • SWE-bench Verified#10 of 16
    69.4%vendorHugging Face: Kwaipilot/KAT-Coder-V2.5-Dev ↗

    Leader: Claude Sonnet 5 at 85.2%

  • SWE-bench Pro#26 of 31
    46%vendorHugging Face: Kwaipilot/KAT-Coder-V2.5-Dev ↗

    Leader: Claude Opus 5.5 at 89.9%

  • Terminal-Bench 2.x#69 of 103
    41%vendorTerminal-Bench 2.1Hugging Face: Kwaipilot/KAT-Coder-V2.5-Dev ↗

    Leader: Claude Fable 5.1 at 91.4%

  • SciCode#58 of 94
    44.2%vendorHugging Face: Kwaipilot/KAT-Coder-V2.5-Dev ↗

    Leader: Claude Opus 5.5 at 66.9%

Other published results

ResultScoreReported bySource
SWE-bench Multilingual63%vendorHugging Face: Kwaipilot/KAT-Coder-V2.5-Dev ↗
PinchBench93.4%vendorHugging Face: Kwaipilot/KAT-Coder-V2.5-Dev ↗
KAT Code Bench46.2%vendorHugging Face: Kwaipilot/KAT-Coder-V2.5-Dev ↗

Specs

Released
Not published
Parameters
35B total, 3B active
Size class
10 to 40B
Input
text
Output
text
Context
262K tokens
Max output
Not published

    Will it fit your budget?

    The planner prices this model against local hardware and other APIs for your workload.

    Plan a stack →

    Compare with

    • KuaishouKAT-Coder-Pro V2.5
    • KuaishouKAT-Coder-Pro V221.7
    • OpenAIGPT-6 Astra52.7
    • OpenAIGPT-6 Sol47.5