Inference
EN

LFM2.5-8B-A1B vs Magistral Small

Change models
LFM2.5-8B-A1BLiquidAI
Magistral SmallMistral
IndexAll categories, out of 100.
30.6 ±7.5
31.4 ±4.3 best
GapIndex points below the top.
−0.8 · no clear gap
Top
CoverageIndex weight that has results.
70%
95% best
SpeedOutput tokens per second.
335/s
—
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2026-05
2025-06
Input / 1MUS dollars per 1M input tokens.
—
—
Output / 1MUS dollars per 1M output tokens.
—
—
1K chats2K tokens in, 500 out, each.
—
—
1K documents20K tokens in, 1K out, each.
—
—
1K agent tasks200K tokens in, 20K out, each.
—
—
ContextMaximum tokens in one request.
128K
128K
AgenticMulti-step tasks with tools.
46.3 best
24.8
CodingCode writing and repair.
—
36.3
ReasoningLogic problems and puzzles.
22.1
28.2 best
MultimodalTasks with images and text.
—
40.0
KnowledgeFacts and expert knowledge.
26.0
31.0 best
MultilingualTasks in many languages.
5.0
—
InstructionTasks with strict prompt rules.
44.8 best
12.2
MathMath problems.
27.1
41.3 best

Benchmarks

9 shared
Only LFM2.5-8B-A1B: 12Only Magistral Small: 20
LFM2.5-8B-A1B
Magistral Small
Artificial Analysis Humanity's Last ExamKnowledge
6.9%163 / 222
7.6% best155 / 222
Artificial Analysis Long Context ReasoningReasoning
0.0%204 / 217
0.0%204 / 217
Critical Physics TasksReasoning
0.0%150 / 214
0.0%150 / 214
Artificial Analysis GPQA DiamondKnowledge
51.3%171 / 199
64.1% best146 / 199
Artificial Analysis Omniscience AccuracyKnowledge
9.4%162 / 164
18.0% best141 / 164
τ²-Bench Tool-Agent-User EvaluationAgentic
88.1% best47 / 149
26.6%116 / 149
Artificial Analysis IFBenchInstruction
55.6% best70 / 146
24.8%141 / 146
American Invitational Mathematics Examination 2025Math
42.5%45 / 53
62.8% best40 / 53
MATH-500 Problem SetMath
88.8%23 / 27
95.9% best15 / 27