Inference
EN

GPT-6.1 Terra vs Step 5 Preview

Change models
GPT-6.1 TerraOpenAI
Step 5 PreviewStepFun
IndexAll categories, out of 100.
77.1 ±3.8 best
68.4 ±3.7
GapIndex points below the top.
Top
−8.7
CoverageIndex weight that has results.
90%
90%
SpeedOutput tokens per second.
46/s
105/s best
EloLMArena rating and rank.
—
1443
ReleasedMonth of release.
2026-10
2026-09
Input / 1MUS dollars per 1M input tokens.
$1.00
$1.00
Output / 1MUS dollars per 1M output tokens.
$5.00
$2.70 best
Batch output / 1MBatch price per 1M output tokens.
$2.50
—
1K chats2K tokens in, 500 out, each.
$4.50
$3.35 best
1K documents20K tokens in, 1K out, each.
$25.00
$22.70 best
1K agent tasks200K tokens in, 20K out, each.
$300.00
$254.00 best
ContextMaximum tokens in one request.
1.05M best
1M
AgenticMulti-step tasks with tools.
78.6 best
68.8
CodingCode writing and repair.
74.4 best
71.4
ReasoningLogic problems and puzzles.
81.3 best
67.4
MultimodalTasks with images and text.
73.2 best
62.2
KnowledgeFacts and expert knowledge.
70.5 best
63.5
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
—
—
MathMath problems.
76.9 best
61.7

Benchmarks

23 shared
Only GPT-6.1 Terra: 27Only Step 5 Preview: 39
GPT-6.1 Terra
Step 5 Preview
Artificial Analysis Humanity's Last ExamKnowledge
46.2%24 / 222
46.5% best23 / 222
Artificial Analysis Long Context ReasoningReasoning
82.7%23 / 217
88.3% best2 / 217
Critical Physics TasksReasoning
28.0% best16 / 214
20.9%26 / 214
Artificial Analysis Omniscience AccuracyKnowledge
47.6% best28 / 164
36.4%64 / 164
Vibe Code Bench v1.1Coding
84.0% best19 / 123
70.4%48 / 123
Artificial Analysis SciCodeCoding
55.1%28 / 114
58.9% best13 / 114
Artificial Analysis MMMU-ProMultimodal
84.7% best6 / 113
76.4%44 / 113
Legal Research BenchLegal
26.0%62 / 90
40.4% best35 / 90
Tax Agent BenchFinance
17.1%58 / 82
27.3% best39 / 82
LiveBench Data AnalysisReasoning
82.3% best4 / 80
63.5%69 / 80
LiveBench ReasoningReasoning
88.7% best20 / 80
83.8%44 / 80
Terminal-Bench 4.0Agentic
49.5% best11 / 61
31.8%23 / 61
Agent Arena steerabilityAgentic
16.6 best4 / 54
−0.930 / 54
Agent Arena task outcomeAgentic
9.7 best11 / 54
−1.535 / 54
Massive Multi-discipline Multimodal Understanding ProMultimodal
85.5% best1 / 53
76.0%33 / 53
LiveBench Coding 2.0Coding
90.8% best11 / 52
85.6%22 / 52
LiveBench DesignAgentic
82.3% best8 / 52
76.9%24 / 52
LiveBench ResearchAgentic
73.9% best25 / 52
72.8%27 / 52
LiveBench Long AgenticAgentic
39.6% best23 / 51
27.5%36 / 51
LiveBench Browser UseAgentic
100.0% best1 / 49
24.0%41 / 49