Claude Sonnet 4.5 Thinking vs GPT-5.4 mini

Change models
IndexAll categories, out of 100.
53.6 ±6.9
54.3 ±2.4 best
GapIndex points below the top.
−0.7 · no clear gap
Top
CoverageIndex weight that has results.
80%
100% best
SpeedOutput tokens per second.
—
93/s
EloLMArena rating and rank.
—
1412
ReleasedMonth of release.
2025-09
2026-03
Input / 1MUS dollars per 1M input tokens.
—
$0.75
Output / 1MUS dollars per 1M output tokens.
—
$4.50
Batch output / 1MBatch price per 1M output tokens.
—
$2.25
1K chats2K tokens in, 500 out, each.
—
$3.75
1K documents20K tokens in, 1K out, each.
—
$19.50
1K agent tasks200K tokens in, 20K out, each.
—
$240.00
ContextMaximum tokens in one request.
200K
400K best
AgenticMulti-step tasks with tools.
56.8 best
55.6
CodingCode writing and repair.
50.4
54.9 best
ReasoningLogic problems and puzzles.
—
51.9
MultimodalTasks with images and text.
58.1 best
54.6
KnowledgeFacts and expert knowledge.
53.1
53.5 best
MultilingualTasks in many languages.
60.2 best
57.4
InstructionTasks with strict prompt rules.
—
52.5
MathMath problems.
53.8 best
53.7

Benchmarks

22 shared
Only Claude Sonnet 4.5 Thinking: 6Only GPT-5.4 mini: 54
Claude Sonnet 4.5 Thinking
GPT-5.4 mini
EuroEval PolishMultilingual
63.8% best10 / 210
62.5%13 / 210
EuroEval SwedishMultilingual
66.0%17 / 196
68.9% best12 / 196
EuroEval GermanMultilingual
57.1%14 / 192
57.8% best12 / 192
64.8% best17 / 183
64.7%19 / 183
EuroEval FrenchMultilingual
66.9%21 / 179
67.4% best16 / 179
EuroEval SpanishMultilingual
60.4%15 / 177
62.8% best9 / 177
EuroEval DutchMultilingual
65.3% best10 / 176
63.1%16 / 176
EuroEval ItalianMultilingual
65.1%21 / 175
66.2% best14 / 175
TaxEval v2Finance
73.3% best50 / 143
71.2%80 / 143
73.0%82 / 142
81.5% best60 / 142
81.6%61 / 135
83.1% best57 / 135
MMLU ProKnowledge
87.4% best28 / 135
84.6%60 / 135
CorpFin v2Finance
62.0% best48 / 131
60.9%62 / 131
22.6%89 / 125
48.0% best72 / 125
64.0% best49 / 97
63.5%52 / 97
AIMEMath
88.2%30 / 95
95.6% best8 / 95
MMMU ProMultimodal
79.3% best48 / 92
79.2%49 / 92
SWE-benchCoding
70.0%61 / 87
73.0% best53 / 87
41.6%29 / 66
44.9% best24 / 66
IOI v1Coding
18.3% best25 / 61
6.4%42 / 61