LiquidAI LFM2.5-2.6B at launch

What Liquid reported at launch

From “LFM2.5-2.6B”, August 11, 2026. The maker's own numbers, not re-run by us.

• IFBench

  59.2

  2.6B

  #16 of 16 models with a reported result

• AA-Omni-Public Acc

  8.1

  2.6B

• AA-Omni-Public Non-hallu

  59

  2.6B

• AIME25

  51.9

  2.6B

  #4 of 4 models with a reported result

• τ³-Banking

  5.7%

  2.6B

  #5 of 5 models with a reported result

• BFCLv4

  56.9

  2.6B

  #2 of 2 models with a reported result

• BrowseComp+ (OpenClaw)

  26.9%

  2.6B

• Claw-Eval average (EN)

  62.9

  2.6B

• IFStruct

  85.5

  2.6B

• LiveCodeBenchv6

  59.4

  2.6B

• Multi-IF

  80.1

  2.6B

  #4 of 4 models with a reported result

• PinchBench

  68.2

  2.6B

  #4 of 4 models with a reported result

Independent results so far

5 scores from 5 boards that test models themselves. Ranks compare LFM2.5-2.6B with every model on the same board.