Sunk Cost sunkcost.ai Data checked 2026-09-03

Llama 3.3 70B Instruct vs LFM2.5 2.6B

Neither has a clear lead on the index. LFM2.5 2.6B is the smaller download at 1.7 GB, so it runs on cheaper hardware.

Llama 3.3 70B InstructLFM2.5 2.6B
Intelligence index88
ClassBelow every hosted tierBelow every hosted tier
Weights43 GB1.7 GB
QuantisationQ4_K_MQ4_K_M
Parameters70.6B2.7B
Max context128k128k
API price per 1M$0.1 in / $0.32 out$0 in / $0 out
LicenceLlama 3.3 Community LicenseLFM 1.0 (see repo)
Cheapest machine that runs itStrix Halo Framework Desktop, 128GB $3,449Mac mini M6, 16GB $899
Summarising good not rated
Translation good not rated
Everyday coding good not rated
Reasoning & maths usable not rated
Agentic work usable not rated

Ratings are coarse on purpose. Speeds and pay-back depend on the machine — open either model's page for the full list, or see both against the frontier. Context is 32k throughout.