Inference
EN

Gemini 4 Carbon vs Step 5 Preview

Change models
Gemini 4 CarbonGoogle
Step 5 PreviewStepFun
IndexAll categories, out of 100.
82.7 ±5.1 best
68.4 ±3.7
GapIndex points below the top.
Top
−14.3
CoverageIndex weight that has results.
80%
90% best
SpeedOutput tokens per second.
31/s
105/s best
EloLMArena rating and rank.
1538 best
1443
ReleasedMonth of release.
2026-10
2026-09
Input / 1MUS dollars per 1M input tokens.
$2.00
$1.00 best
Output / 1MUS dollars per 1M output tokens.
$10.00
$2.70 best
Batch output / 1MBatch price per 1M output tokens.
$5.00
—
1K chats2K tokens in, 500 out, each.
$9.00
$3.35 best
1K documents20K tokens in, 1K out, each.
$50.00
$22.70 best
1K agent tasks200K tokens in, 20K out, each.
$600.00
$254.00 best
ContextMaximum tokens in one request.
1.05M best
1M
AgenticMulti-step tasks with tools.
83.2 best
68.8
CodingCode writing and repair.
80.5 best
71.4
ReasoningLogic problems and puzzles.
83.9 best
67.4
MultimodalTasks with images and text.
—
62.2
KnowledgeFacts and expert knowledge.
79.5 best
63.5
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
—
—
MathMath problems.
77.2 best
61.7

Benchmarks

11 shared
Only Gemini 4 Carbon: 28Only Step 5 Preview: 51
Gemini 4 Carbon
Step 5 Preview
Artificial Analysis Long Context ReasoningReasoning
79.3%52 / 217
88.3% best2 / 217
Artificial Analysis Omniscience AccuracyKnowledge
58.3% best9 / 164
36.4%64 / 164
Vibe Code Bench v1.1Coding
89.1% best9 / 123
70.4%48 / 123
Legal Research BenchLegal
55.8% best1 / 90
40.4%35 / 90
Tax Agent BenchFinance
51.8% best1 / 82
27.3%39 / 82
Terminal-Bench 4.0Agentic
63.1% best4 / 61
31.8%23 / 61
Agent Arena steerabilityAgentic
19.1 best2 / 54
−0.930 / 54
Agent Arena task outcomeAgentic
18.4 best4 / 54
−1.535 / 54
HLAB-Vals-1.1Legal
17.5% best8 / 35
12.9%15 / 35
AutomationBenchAgentic
53.9% best5 / 32
44.0%17 / 32
Multimodal Knowledge Agent BenchmarkAgentic
54.3% best2 / 19
27.6%19 / 19