Inference
EN

Gemini 3.8 Flash vs Step 5 Preview

Change models
Gemini 3.8 FlashGoogle
Step 5 PreviewStepFun
IndexAll categories, out of 100.
67.8 ±2.5
68.4 ±3.7 best
GapIndex points below the top.
−0.6 · no clear gap
Top
CoverageIndex weight that has results.
95% best
90%
SpeedOutput tokens per second.
68/s
105/s best
EloLMArena rating and rank.
1500 best
1443
ReleasedMonth of release.
2026-09
2026-09
Input / 1MUS dollars per 1M input tokens.
$0.825 best
$1.00
Output / 1MUS dollars per 1M output tokens.
$4.13
$2.70 best
Batch output / 1MBatch price per 1M output tokens.
$1.88
—
1K chats2K tokens in, 500 out, each.
$3.71
$3.35 best
1K documents20K tokens in, 1K out, each.
$20.63 best
$22.70
1K agent tasks200K tokens in, 20K out, each.
$247.50 best
$254.00
ContextMaximum tokens in one request.
1.05M best
1M
AgenticMulti-step tasks with tools.
65.3
68.8 best
CodingCode writing and repair.
66.2
71.4 best
ReasoningLogic problems and puzzles.
68.3 best
67.4
MultimodalTasks with images and text.
67.1 best
62.2
KnowledgeFacts and expert knowledge.
70.9 best
63.5
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
91.3
—
MathMath problems.
64.3 best
61.7

Benchmarks

42 shared
Only Gemini 3.8 Flash: 64Only Step 5 Preview: 20
Gemini 3.8 Flash
Step 5 Preview
Artificial Analysis Humanity's Last ExamKnowledge
47.8% best20 / 222
46.5%23 / 222
Artificial Analysis Long Context ReasoningReasoning
81.3%34 / 217
88.3% best2 / 217
Critical Physics TasksReasoning
18.3%34 / 214
20.9% best26 / 214
Artificial Analysis Omniscience AccuracyKnowledge
50.5% best19 / 164
36.4%64 / 164
GDPval-AA normalizedAgentic
46.8%34 / 133
54.3% best24 / 133
Vibe Code Bench v1.1Coding
78.7% best34 / 123
70.4%48 / 123
Artificial Analysis SciCodeCoding
56.6%23 / 114
58.9% best13 / 114
Artificial Analysis MMMU-ProMultimodal
85.6% best4 / 113
76.4%44 / 113
Legal Research BenchLegal
38.9%38 / 90
40.4% best35 / 90
Terminal-Bench 2.1 · BenchLMAgentic
89.4% best5 / 86
85.0%20 / 86
Tax Agent BenchFinance
32.1% best23 / 82
27.3%39 / 82
LiveBench Data AnalysisReasoning
54.0%76 / 80
63.5% best69 / 80
LiveBench ReasoningReasoning
89.3% best18 / 80
83.8%44 / 80
Finance Agent (v2)Finance
61.4% best2 / 79
50.7%42 / 79
Code MigrationCoding
36.5%35 / 77
39.0% best31 / 77
EMBFinance
72.2% best8 / 75
57.1%41 / 75
Terminal-Bench 4.0Agentic
19.7%34 / 61
31.8% best23 / 61
DeepSWEAgentic
73.8% best12 / 59
67.7%27 / 59
Agent Arena steerabilityAgentic
−1.333 / 54
−0.9 best30 / 54
Agent Arena task outcomeAgentic
1.3 best29 / 54
−1.535 / 54