833
source domains used, nearly 9x the best frontier model
1,062
competitive-set entities mapped, 3 to 8x more than frontier models
~30%
of figures from the frontier models were unusable in practice
01
Source domains used on the same task
Caliper Lab ran an identical commercial due diligence brief through Binocs and three leading AI models, then counted the distinct source domains each one actually drew on.
833
96
64
81
Binocs
Claude Opus 4.8
GPT-5.5
GPT Deep Research
Unique citation domains, six reports per system. Source: Caliper Lab, June 2026.
What the gap looks like in practice
More sources means a wider view of the competitive set: Caliper Lab found Binocs mapped 1,062 entities in the same market, 3 to 8x more than the frontier models in the test. It also found that about 30% of the figures the frontier models produced were unusable in practice: not sourced, not reconcilable, or outright fabricated. Binocs’ output was model-ready across the board.
