LFM3-1.7B
- Input
- —
- Output
- —
- Speed
- ~120tok/sest.
- Context
- 262Ktokens
Liquid AI's LFM Open License v1.0 small model for local agents (hybrid conv/attention, 256K context). Card also reports BFCL V4 71.39, ToolSandbox 80.36, LiveCodeBench v6 73.46 and RULER 128K 65.56, left out (not day-0 benchmarks here). OpenRouter lists $0.08/$0.20 but that is not an AA host median, so no price is entered.
Rankings
Self-hostingEstimated
How we estimate →What it would cost to run these open weights yourself on rented GPUs. No API sells this model, so this is its price on the chart.
- Hardware
- 1× B200 180GB
- BF16 weights
- Throughput
- ~24,713 tok/s
- many requests batched
- Price
- $0.028–0.25 /M tok
- busy → light use
- On your own machine
- RTX 5090 32GB (int4, 300+ tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- BF16 weights (1.73B params) + 25% KV-cache headroom ≈ 4 GB
- 1× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
- ~24,713 output tok/s aggregate at batch 256 (bandwidth-bound)
- Blended 3:1 input:output; prefill ~307,262 tok/s
- Low = 75% utilization at the low GPU price; high = 20% at the high price
Benchmark scores
~ italic, dashed = no published score yet, estimated from related benchmarks.
Writing & chat
IFBenchLab-reported
#29 of 13973.0%
- #1MiniMax-M382.9%
- #2Qwen3.8 2.4T A95B82.8%
- #28GPT-5.4 mini73.3%
- #29LFM3-1.7B73.0%
- #30Gemma 4 26B A4B72.4%
huggingface.co · 2026-11-09
LMArena Text, Non-English (Elo)Estimated
~#37 of 57 (estimated)~1410 Elo
- #1Gemini 4 ArgonFew votes1512
- #2Claude Opus 5.51497
- #36MiMo-V2.51411
- ~#37LFM3-1.7BEstimated1410
- #37Mistral Medium 3.51406
No published score — estimated from related benchmarks
LMArena Text (Elo)Estimated
~#38 of 57 (estimated)~1425 Elo
- #1Gemini 4 Argon1525
- #2Claude Opus 5.51507
- #36Nemotron 3 Ultra1426
- ~#38LFM3-1.7BEstimated1425
- #38Muse GlimmerFew votes1424
No published score — estimated from related benchmarks
EQ-Bench Creative Writing v3 (Elo)Estimated
~#15 of 26 (estimated)~1793 Elo
- #1GPT-6 Astra2173
- #2Claude Fable 5.12162
- #14Claude Sonnet 51794
- ~#15LFM3-1.7BEstimated1793
- #15Gemini 3.8 Flash1748
No published score — estimated from related benchmarks
Hard reasoning
GPQA DiamondLab-reported
#129 of 19263.4%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #128LongCat Flash Lite63.6%
- #129LFM3-1.7B63.4%
- #130Sarvam 30B63.3%
huggingface.co · 2026-11-09
Humanity's Last ExamEstimated
~#90 of 199 (estimated)~16.5%
- #1Claude Fable 5.274.3%
- #2Claude Opus 5.561.4%
- #89HyperNova 60B 260516.6%
- ~#90LFM3-1.7BEstimated16.5%
- #90KAT Coder Pro V216.1%
No published score — estimated from related benchmarks
AIME (latest)Estimated
~#23 of 34 (estimated)~56.1%
- #1GLM-5.3 NVFP499.5%
- #2Beam 1.0 501B A23B97.7%
- #22Phi-4-reasoning-plus78.0%
- ~#23LFM3-1.7BEstimated56.1%
- #23Mistral Large 338.0%
No published score — estimated from related benchmarks
AA Intelligence IndexEstimated
~#72 of 103 (estimated)~15.0
- #1Gemini 4 Carbon59.4
- #2Claude Opus 5.557.6
- #70Qwen3.5 122B A10B15.6
- ~#72LFM3-1.7BEstimated15.0
- #72Gemma 4 31B14.7
No published score — estimated from related benchmarks