Head-to-head comparison across 1benchmark categories. Overall scores shown here use BenchLM's provisional ranking lane.
Composer 2.5
64
GPT-5.2
79
Pick GPT-5.2 if you want the stronger benchmark profile. Composer 2.5 only becomes the better choice if agentic is the priority or you want the cheaper token bill.
Agentic
+14.1 difference
Composer 2.5
GPT-5.2
$0.5 / $2.5
$1.75 / $14
N/A
73 t/s
N/A
130.34s
200K
400K
Pick GPT-5.2 if you want the stronger benchmark profile. Composer 2.5 only becomes the better choice if agentic is the priority or you want the cheaper token bill.
GPT-5.2 is clearly ahead on the provisional aggregate, 79 to 64. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
GPT-5.2 is also the more expensive model on tokens at $1.75 input / $14.00 output per 1M tokens, versus $0.50 input / $2.50 output per 1M tokens for Composer 2.5. That is roughly 5.6x on output cost alone. GPT-5.2 gives you the larger context window at 400K, compared with 200K for Composer 2.5.
GPT-5.2 is ahead on BenchLM's provisional leaderboard, 79 to 64.
Composer 2.5 has the edge for agentic tasks in this comparison, averaging 69.3 versus 55.2. GPT-5.2 stays close enough that the answer can still flip depending on your workload.
For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.
Free. No spam. Unsubscribe anytime.