GPT-5 (medium) vs Mercury 2

Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.

GPT-5 (medium) is clearly ahead on the aggregate, 78 to 65. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.

GPT-5 (medium)'s sharpest advantage is in coding, where it averages 66.1 against 41.1. The single biggest benchmark swing on the page is SWE-bench Pro, 72 to 43.

Quick Verdict

Pick GPT-5 (medium) if you want the stronger benchmark profile. Mercury 2 only becomes the better choice if its workflow or ecosystem matters more than the raw scoreboard.

Agentic

GPT-5 (medium)

GPT-5 (medium)

75.5

Mercury 2

63.7

77
Terminal-Bench 2.0
63
78
BrowseComp
67
72
OSWorld-Verified
62

Coding

GPT-5 (medium)

GPT-5 (medium)

66.1

Mercury 2

41.1

83
HumanEval
75
67
SWE-bench Verified
46
60
LiveCodeBench
38
72
SWE-bench Pro
43

Multimodal & Grounded

GPT-5 (medium)

GPT-5 (medium)

88.1

Mercury 2

68.3

89
MMMU-Pro
66
87
OfficeQA Pro
71

Reasoning

GPT-5 (medium)

GPT-5 (medium)

84.3

Mercury 2

80.1

87
SimpleQA
82
85
MuSR
82
92
BBH
87
81
LongBench v2
77
81
MRCRv2
76

Knowledge

GPT-5 (medium)

GPT-5 (medium)

69.8

Mercury 2

57.2

91
MMLU
78
89
GPQA
78
87
SuperGPQA
76
85
OpenBookQA
74
81
MMLU-Pro
72
27
HLE
9
82
FrontierScience
69

Instruction Following

GPT-5 (medium)

GPT-5 (medium)

88

Mercury 2

84

88
IFEval
84

Multilingual

GPT-5 (medium)

GPT-5 (medium)

88.1

Mercury 2

79.7

90
MGSM
81
87
MMLU-ProX
79

Mathematics

GPT-5 (medium)

GPT-5 (medium)

92

Mercury 2

80.9

93
AIME 2023
81
95
AIME 2024
83
94
AIME 2025
82
89
HMMT Feb 2023
77
91
HMMT Feb 2024
79
90
HMMT Feb 2025
78
92
BRUMO 2025
80
92
MATH-500
82

Frequently Asked Questions

Which is better, GPT-5 (medium) or Mercury 2?

GPT-5 (medium) is ahead overall, 78 to 65. The biggest single separator in this matchup is SWE-bench Pro, where the scores are 72 and 43.

Which is better for knowledge tasks, GPT-5 (medium) or Mercury 2?

GPT-5 (medium) has the edge for knowledge tasks in this comparison, averaging 69.8 versus 57.2. Inside this category, HLE is the benchmark that creates the most daylight between them.

Which is better for coding, GPT-5 (medium) or Mercury 2?

GPT-5 (medium) has the edge for coding in this comparison, averaging 66.1 versus 41.1. Inside this category, SWE-bench Pro is the benchmark that creates the most daylight between them.

Which is better for math, GPT-5 (medium) or Mercury 2?

GPT-5 (medium) has the edge for math in this comparison, averaging 92 versus 80.9. Inside this category, AIME 2023 is the benchmark that creates the most daylight between them.

Which is better for reasoning, GPT-5 (medium) or Mercury 2?

GPT-5 (medium) has the edge for reasoning in this comparison, averaging 84.3 versus 80.1. Inside this category, SimpleQA is the benchmark that creates the most daylight between them.

Which is better for agentic tasks, GPT-5 (medium) or Mercury 2?

GPT-5 (medium) has the edge for agentic tasks in this comparison, averaging 75.5 versus 63.7. Inside this category, Terminal-Bench 2.0 is the benchmark that creates the most daylight between them.

Which is better for multimodal and grounded tasks, GPT-5 (medium) or Mercury 2?

GPT-5 (medium) has the edge for multimodal and grounded tasks in this comparison, averaging 88.1 versus 68.3. Inside this category, MMMU-Pro is the benchmark that creates the most daylight between them.

Which is better for instruction following, GPT-5 (medium) or Mercury 2?

GPT-5 (medium) has the edge for instruction following in this comparison, averaging 88 versus 84. Inside this category, IFEval is the benchmark that creates the most daylight between them.

Which is better for multilingual tasks, GPT-5 (medium) or Mercury 2?

GPT-5 (medium) has the edge for multilingual tasks in this comparison, averaging 88.1 versus 79.7. Inside this category, MGSM is the benchmark that creates the most daylight between them.

Last updated: March 12, 2026

Weekly LLM Benchmark Digest

Get notified when new models drop, benchmark scores change, or the leaderboard shifts. One email per week.

Free. No spam. Unsubscribe anytime. We only store derived location metadata for consent routing.