ÆAAI
← All models
GLM51Z.ai · GLMSynthesizedOpen Weights

Z.ai GLM-5.1

Z.ai's next-gen open-weights flagship —

Accounting overall

71.3%

Input / Output

$1.40 / $4.40 per MTok

Context

200K

Speed

~110 tok/s

Released

2026-04

Cutoff

2026-01

GDPVal-AA Elo

1535

01Accounting Task Breakdown

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

Transaction Class.
85.0
Journal Entry
83.0
Accounts Payable
75.0
Accounts Receivable
73.0
Bank Reconciliation
71.0
Financial Reporting
57.0
Month-End Close
45.0
Accounting Knowledge
81.0
02Research

GLM-5.1 released April 7, 2026 as a 754B-parameter open-weights model under MIT license. It is the first open-weights model on Artificial Analysis' GDPVal-AA leaderboard above 1500 Elo (1,535, #8 overall), beating every Chinese peer and every Gemini variant currently ranked. On SWE-Bench Pro it scores 58.4, edging GPT-5.4 (57.7) and Claude Opus 4.6 (57.3). Z.ai positions it for long-horizon autonomous multi-hour agentic loops.

Accounting sub-scores are **synthesized** — not yet on DualEntry. We estimate ~73% overall by extrapolating from GLM-5's measured 72.3% plus GLM-5.1's meaningful gains on coding and agentic benchmarks. Predecessor GLM-5 is still live on Z.ai but Nvidia NIM flagged a deprecation for April 20, 2026; GLM-5.1 is the forward-compatible path.

For accounting tools that already integrate GLM-5, the migration is low-risk and the uplift on multi-step agentic workflows (close, reporting) should be real. Pricing is up materially vs GLM-5 ($0.60/$1.80 → $1.40/$4.40), so the cost-per-token calculus is tighter. The open-weights release sidesteps data-sovereignty concerns for enterprise buyers who want to self-host.

Citations