Claude Haiku 4.5 Thinking vs GPT-OSS 120B

Change models
IndexAll categories, out of 100.
43.3 ±7.0 best
43.2 ±4.2
GapIndex points below the top.
Top
−0.1 · no clear gap
CoverageIndex weight that has results.
75%
90% best
SpeedOutput tokens per second.
—
224/s
EloLMArena rating and rank.
—
1365
ReleasedMonth of release.
2025-10
2025-08
Input / 1MUS dollars per 1M input tokens.
—
$0.09
Output / 1MUS dollars per 1M output tokens.
—
$0.36
1K chats2K tokens in, 500 out, each.
—
$0.36
1K documents20K tokens in, 1K out, each.
—
$2.16
1K agent tasks200K tokens in, 20K out, each.
—
$25.20
ContextMaximum tokens in one request.
200K best
131K
AgenticMulti-step tasks with tools.
49.8 best
40.1
CodingCode writing and repair.
42.4
42.6 best
ReasoningLogic problems and puzzles.
—
40.1
MultimodalTasks with images and text.
30.4
—
KnowledgeFacts and expert knowledge.
43.3 best
42.5
MultilingualTasks in many languages.
—
51.0
InstructionTasks with strict prompt rules.
—
52.8
MathMath problems.
51.1
51.2 best

Benchmarks

14 shared
Only Claude Haiku 4.5 Thinking: 14Only GPT-OSS 120B: 34
Claude Haiku 4.5 Thinking
GPT-OSS 120B
81.2% best82 / 149
75.9%113 / 149
TaxEval v2Finance
67.5%106 / 143
71.6% best76 / 143
41.2%129 / 142
83.2% best49 / 142
72.2%88 / 135
78.5% best72 / 135
MMLU ProKnowledge
78.7%100 / 135
79.2% best98 / 135
CorpFin v2Finance
60.6% best66 / 131
58.2%84 / 131
AIMEMath
82.7%44 / 95
92.6% best18 / 95
MedQAHealthcare
79.6%74 / 95
91.4% best38 / 95
SWE-benchCoding
66.6% best70 / 87
33.6%86 / 87
MGSMMultilingual
92.1% best21 / 74
92.0%24 / 74
38.2% best36 / 66
19.1%58 / 66
56.5% best30 / 53
48.8%49 / 53
46.9% best32 / 51
21.5%45 / 51
50.0% best10 / 47
22.5%37 / 47