Refresh Models —Deals —Regulatory —Sources —

LFM2.5-1.2B-Thinking vs LFM2.5-1.2B-Instruct

21 SHARED BENCHMARKS

Across 21 shared benchmarks, LFM2.5-1.2B-Thinking scores higher on 12 and LFM2.5-1.2B-Instruct on 5, with 4 level. The widest gap is OmniScience Non-Hallucination, where LFM2.5-1.2B-Instruct scores 14.9 against 5.6.

LIQUID AIVSLIQUID AI21 SHARED12–5 HEAD-TO-HEAD

BenchmarkLFM2.5-1.2B-ThinkingMARGINLFM2.5-1.2B-Instruct

AA Agentic Index6.53.6

AA Intelligence2.32.3

AA-Omniscience-80-72.1

AIME25 no tools31.714

Artificial Analysis Coding Index1.40.8

GPQA Diamond37.938.9

GSM8K85.664.5

HLE6.26.7

IFBench44.947.3

ifeval88.486.2

math_500_em8863.2

MMLU-Pro49.644.4

OmniScience Accuracy7.47

OmniScience Non-Hallucination5.614.9

Terminal-Bench Hard00

τ²-Bench Telecom (AA run)19.610.8

Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.