Refresh Models —Deals —Regulatory —Sources —

LFM2.5-1.2B-Instruct vs LFM2.5-8B-A1B

23 SHARED BENCHMARKS

Across 23 shared benchmarks, LFM2.5-1.2B-Instruct scores higher on 1 and LFM2.5-8B-A1B on 19, with 3 level. The widest gap is Artificial Analysis Coding Index, where LFM2.5-8B-A1B scores 5.6 against 0.8.

LIQUID AIVSLIQUID AI23 SHARED1–19 HEAD-TO-HEAD

BenchmarkLFM2.5-1.2B-InstructMARGINLFM2.5-8B-A1B

AA Agentic Index3.65.4

AA Intelligence2.38.3

AA-Omniscience-72.1-33.2

AIME25 no tools1442.5

Artificial Analysis Coding Index0.85.6

BFCL v349.164.8

GPQA Diamond38.951.3

GSM8K64.54

HLE6.76.9

IFBench47.356.5

ifeval86.291.8

MATH500 (Pass@1)688.8

MBPP5.56.9

OmniScience Accuracy79.3

OmniScience Non-Hallucination14.953.1

Terminal-Bench Hard04.5

τ²-Bench Telecom (AA run)10.816.1

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.