ÆAAI
← All models
KIMI26 · Kimi K2SynthesizedOpen Weights

Kimi K2.6

Moonshot's 1T open-weights MoE — AA Intelligence Index 54, top open-model on HLE with tools.

Accounting overall

68.1%

Input / Output

$0.95 / $4.00 per MTok

Context

256K

Speed

~134 tok/s

Released

2026-04

Cutoff

2026-02

01Accounting Task Breakdown

Eight accounting-task categories borrowed from DualEntry's 101-task benchmark. Measured where published, synthesized from adjacent benchmarks otherwise.

Transaction Class.
82.0
Journal Entry
80.0
Accounts Payable
72.0
Accounts Receivable
71.0
Bank Reconciliation
66.0
Financial Reporting
54.0
Month-End Close
42.0
Accounting Knowledge
78.0
02Research

Kimi K2.6 was released April 20, 2026 by Moonshot AI. It is a 1T-parameter MoE (32B active) open-weights model with 256K context. On Artificial Analysis' Intelligence Index it scores 54 — ranked #1 of 72 open-weights models and within striking distance of GPT-5.4. On Humanity's Last Exam with tools it hits 54.0, beating GPT-5.4 (52.1) and matching Claude Opus 4.6. SWE-Bench Pro: 58.6 (vs. GPT-5.4 57.7, Opus 4.6 53.4).

Accounting task sub-scores here are **synthesized** — DualEntry has not yet run K2.6 on the 101-task suite. The estimate (~71% overall) places K2.6 between MiniMax M2.7 (71.3% measured) and GLM-5 (72.3% measured) based on adjacent benchmarks and AA Index positioning. Update with measured scores when DualEntry publishes.

For agentic accounting teams with data-residency constraints, K2.6's open-weights release plus 300-parallel-sub-agent orchestration is a legitimate alternative to hosted frontier APIs. Caveats: no native DualEntry score, knowledge cutoff ~February 2026 so Q1 2026 accounting-standards updates may be absent, and the $0.95/$4.00 official pricing is noticeably higher than peer Chinese open-weights models.

Citations