Inference
EN

Gemini 4 Carbon vs GPT-5.6 Sol

Change models
Gemini 4 CarbonGoogle
GPT-5.6 SolOpenAI
IndexAll categories, out of 100.
82.7 ±5.1 best
73.0 ±1.9
GapIndex points below the top.
Top
−9.7
CoverageIndex weight that has results.
80%
100% best
SpeedOutput tokens per second.
31/s
64/s best
EloLMArena rating and rank.
1538 best
1457
ReleasedMonth of release.
2026-10
2026-07
Input / 1MUS dollars per 1M input tokens.
$2.00 best
$4.40
Output / 1MUS dollars per 1M output tokens.
$10.00 best
$22.00
Batch output / 1MBatch price per 1M output tokens.
$5.00 best
$10.00
1K chats2K tokens in, 500 out, each.
$9.00 best
$19.80
1K documents20K tokens in, 1K out, each.
$50.00 best
$110.00
1K agent tasks200K tokens in, 20K out, each.
$600.00 best
$1,320.00
ContextMaximum tokens in one request.
1.05M
1.05M
AgenticMulti-step tasks with tools.
83.2 best
71.9
CodingCode writing and repair.
80.5 best
70.4
ReasoningLogic problems and puzzles.
83.9 best
76.3
MultimodalTasks with images and text.
—
72.6
KnowledgeFacts and expert knowledge.
79.5 best
72.7
MultilingualTasks in many languages.
—
66.4
InstructionTasks with strict prompt rules.
—
64.5
MathMath problems.
77.2 best
75.1

Benchmarks

27 shared
Only Gemini 4 Carbon: 12Only GPT-5.6 Sol: 98
Gemini 4 Carbon
GPT-5.6 Sol
Artificial Analysis Long Context ReasoningReasoning
79.3%52 / 217
84.0% best8 / 217
GPQA diamond · Epoch AIKnowledge
96.5% best1 / 217
93.5%17 / 217
OTIS Mock AIME 2024-2025Math
100.0%1 / 200
100.0%1 / 200
Artificial Analysis Omniscience AccuracyKnowledge
58.3%9 / 164
59.4% best8 / 164
ARC-AGI-2 (semi-private)Reasoning
92.1%5 / 131
92.5% best3 / 131
ARC-AGI-1 (semi-private)Reasoning
97.5% best5 / 126
96.5%11 / 126
Vibe Code Bench v1.1Coding
89.1% best9 / 123
80.5%31 / 123
ARC-AGI-2 (public eval)Reasoning
97.9% best3 / 122
93.8%10 / 122
ARC-AGI-1 (public eval)Reasoning
98.5%6 / 116
99.0% best2 / 116
SimpleQA VerifiedKnowledge
77.7% best2 / 91
69.7%10 / 91
Chess PuzzlesReasoning
78.0% best2 / 90
59.0%10 / 90
FrontierMath-Tiers-1-3-v2-PrivateMath
90.5% best6 / 90
89.1%9 / 90
Legal Research BenchLegal
55.8% best1 / 90
48.1%12 / 90
Tax Agent BenchFinance
51.8% best1 / 82
31.1%29 / 82
Mystery Game PuzzlesReasoning
73.0% best5 / 80
58.0%10 / 80
FrontierMath-Tier-4-v2-PrivateMath
95.1% best4 / 72
82.9%10 / 72
Terminal-Bench 4.0Agentic
63.1% best4 / 61
37.9%19 / 61
Agent Arena steerabilityAgentic
19.1 best2 / 54
5.516 / 54
Agent Arena task outcomeAgentic
18.4 best4 / 54
2.325 / 54
Furniture AssemblyReasoning
77.5% best6 / 37
56.7%12 / 37