- LiveClaw (OpenClaw agent, assumed to be LiveClawBench) 45.6
- Claw-Eval General (EN), average 68.6
- Claw-Eval General (EN), pass^3 51.4
- BrowseComp-Plus (OpenClaw subset) 37
- WebArena-Verified 38.4
- EnterpriseOps-Gym 28.2
- AutomationBench (public) 41.3
- BFCL V4 71.4
- ToolSandbox 80.4
- MCP-Atlas 32.6
- τ³-bench 68
- SWE-bench Verified (OpenCode) 43.4
- SWE-bench Multilingual (OpenCode) 23
- Terminal-Bench 2.1 (OpenCode) 12.9
- LiveCodeBench v6 73.5
- LiveCodeBench Pro 25Q2 31.7
- SciCode 28.5
- IFEval 96.7
- IFBench 73
- IFStruct 93.7
- Multi-IF 89.1
- GPQA Diamond 63.4
- Best of Humanity's Last Exam no tools 7.17 / with tools 13.90 13.9
- AA-Omniscience (public) index 0.50, accuracy 7.07, non-hallucination 92.93 0.5
- AA-LCR 52.8
- RULER 128K 65.6
LFM3-1.7B
LFM3-1.7B ranks first on Multi-IF.
24
Benchmarks with a score
1
Ranked first, out of five or more models
2
In the top three
Liquid AI Oct 2026
LFM3-1.7B
Open weights Flagship
No Epoch Capabilities Index score yet
- IFBench 73% #29 /70
- SciCode 28.5% #44 /47
- AA-LCR (Artificial Analysis Long Context Reasoning) 52.8% #25 /31
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Multi-IF | 89.1% | #1 of 28 | Model card | |
| IFEval | 96.7% | #5 of 126 | Model card | |
| IFBench | 73% | #29 of 70 | Model card | |
| IFStruct | 93.7% | #1 of 2 | Model card | |
| AA-LCR (Artificial Analysis Long Context Reasoning) | 52.8% | #25 of 31 | Model card | |
| Humanity's Last Exam | 13.9% | #120 of 147 | Model card | |
| RULER | 65.6% | #36 of 38 | Model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 63.4% | #200 of 284 | Model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v6 | 73.5% | #44 of 76 | Model card | |
| SWE-bench Verified | 43.4% | #134 of 150 | Model card | |
| SciCode | 28.5% | #44 of 47 | Model card | |
| SWE-bench Multilingual | 23% | #70 of 70 | Model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BFCL V4 | 71.4% | #7 of 24 | Model card | |
| ToolSandbox | 80.4% | #2 of 5 | Model card | |
| AutomationBench | 41.3% | #17 of 41 | Model card | |
| Claw-Eval | 68.6% | #7 of 14 | Model card | |
| tau3-bench | 68% | #6 of 12 | Model card | |
| BrowseComp-Plus | 37% | #3 of 4 | Model card | |
| Claw-Eval Pass^3 | 51.4% | #9 of 11 | Model card | |
| WebArena-Verified | 38.4% | #5 of 6 | Model card | |
| MCP Atlas | 32.6% | #39 of 40 | Model card | |
| Terminal-Bench 2.1 | 12.9% | #53 of 53 | Model card | |
| EnterpriseOps-Gym | 28.2% | #2 of 2 | Model card | |
| LiveClawBench | 45.6% | #2 of 2 | Model card |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.