Head-to-head comparison across 3 benchmark categories
Claude Mythos Preview
84
GPT-5.4 nano
58
Pick Claude Mythos Preview if you want the stronger benchmark profile. GPT-5.4 nano only becomes the better choice if you want the cheaper token bill.
Agentic
+38.0 difference
Knowledge
+37.2 difference
Multimodal
+26.6 difference
Claude Mythos Preview
GPT-5.4 nano
$25 / $125
$0.2 / $1.25
N/A
191 t/s
N/A
3.64s
1M
400K
Pick Claude Mythos Preview if you want the stronger benchmark profile. GPT-5.4 nano only becomes the better choice if you want the cheaper token bill.
Claude Mythos Preview is clearly ahead on the aggregate, 84 to 58. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
Claude Mythos Preview's sharpest advantage is in agentic, where it averages 80.9 against 42.9. The single biggest benchmark swing on the page is OSWorld-Verified, 79.6% to 39%.
Claude Mythos Preview is also the more expensive model on tokens at $25.00 input / $125.00 output per 1M tokens, versus $0.20 input / $1.25 output per 1M tokens for GPT-5.4 nano. That is roughly 100.0x on output cost alone. Claude Mythos Preview gives you the larger context window at 1M, compared with 400K for GPT-5.4 nano.
Claude Mythos Preview is ahead overall, 84 to 58. The biggest single separator in this matchup is OSWorld-Verified, where the scores are 79.6% and 39%.
Claude Mythos Preview has the edge for knowledge tasks in this comparison, averaging 74.9 versus 37.7. Inside this category, HLE w/o tools is the benchmark that creates the most daylight between them.
Claude Mythos Preview has the edge for agentic tasks in this comparison, averaging 80.9 versus 42.9. Inside this category, OSWorld-Verified is the benchmark that creates the most daylight between them.
Claude Mythos Preview has the edge for multimodal and grounded tasks in this comparison, averaging 92.7 versus 66.1. Inside this category, MMMU-Pro is the benchmark that creates the most daylight between them.
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.