Benchmark
First measurement: 27 models read 100 real invoices
The winner costs $0.0009 per invoice and misses none out of a hundred. The two most expensive models in the panel miss one — and three models invent a VAT number that does not exist.
Hub news
What enters the rankings, what changes in the method, and the reasoning behind every choice.
Featured
Announcement
The hub opens up to ten business functions
From invoice reading alone to 22 benchmarks, from finance to management. One is measured today; the other twenty-one have a written protocol and a test set still to be built.
Analysis
Four measures, never a single score
Accuracy, cost, time, hallucinations: the hub never melts them into one number. A model can top the accuracy table and still be unusable because it makes things up.
All articles
24 articles
Benchmark
Model
Qwen3.8 Max Prime evaluated on the hub's only benchmark
Written from the rankings
Model
Claude Opus 5.5 evaluated on the hub's only benchmark
Written from the rankings
Model
Command A+ evaluated on the hub's only benchmark
Written from the rankings
Model
GPT-6 Luna evaluated on the hub's only benchmark
Written from the rankings
Model
GPT-6 Sol evaluated on the hub's only benchmark
Written from the rankings
Model
Grok 4.7 evaluated on the hub's only benchmark
Written from the rankings
Announcement
Model
DeepSeek V4.1 Flash evaluated on the hub's only benchmark
Written from the rankings
Analysis
Model
GPT-6 Astra evaluated on the hub's only benchmark
Written from the rankings
Model
Qwen3.8 Max (0902) evaluated on the hub's only benchmark
Written from the rankings
Model
Gemini 3.8 Flash evaluated on the hub's only benchmark
Written from the rankings
Model
Muse Spark 1.3 evaluated on the hub's only benchmark
Written from the rankings
Model
Claude Fable 5.1 evaluated on the hub's only benchmark
Written from the rankings
Analysis
Model
GLM 5.3 Flash evaluated on the hub's only benchmark
Written from the rankings
Analysis
Benchmark
Model
Qwen3.8 27B evaluated on the hub's only benchmark
Written from the rankings
Model
Qwen3.7 Flash evaluated on the hub's only benchmark
Written from the rankings
Model
Gemini 3.5 Flash Lite evaluated on the hub's only benchmark
Written from the rankings
Model
Kimi K3 evaluated on the hub's only benchmark
Written from the rankings
Model
Claude Sonnet 5 evaluated on the hub's only benchmark
Written from the rankings