Head-to-head comparison across 2benchmark categories. Overall scores shown here use BenchLM's provisional ranking lane.
LongCat-2.0
80
Step 3.7 Flash
67
Pick LongCat-2.0 if you want the stronger benchmark profile. Step 3.7 Flash only becomes the better choice if you want the cheaper token bill.
Agentic
+4.9 difference
Coding
+3.2 difference
LongCat-2.0
Step 3.7 Flash
$0.75 / $2.95
$0.2 / $1.15
N/A
N/A
N/A
N/A
1M
256K
Pick LongCat-2.0 if you want the stronger benchmark profile. Step 3.7 Flash only becomes the better choice if you want the cheaper token bill.
LongCat-2.0 is clearly ahead on the provisional aggregate, 80 to 67. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
LongCat-2.0's sharpest advantage is in agentic, where it averages 70.8 against 65.9. The single biggest benchmark swing on the page is Terminal-Bench 2.0, 70.8% to 59.5%.
LongCat-2.0 is also the more expensive model on tokens at $0.75 input / $2.95 output per 1M tokens, versus $0.20 input / $1.15 output per 1M tokens for Step 3.7 Flash. That is roughly 2.6x on output cost alone. LongCat-2.0 gives you the larger context window at 1M, compared with 256K for Step 3.7 Flash.
LongCat-2.0 is ahead on BenchLM's provisional leaderboard, 80 to 67. The biggest single separator in this matchup is Terminal-Bench 2.0, where the scores are 70.8% and 59.5%.
LongCat-2.0 has the edge for coding in this comparison, averaging 59.5 versus 56.3. Inside this category, Terminal-Bench 2.0 is the benchmark that creates the most daylight between them.
LongCat-2.0 has the edge for agentic tasks in this comparison, averaging 70.8 versus 65.9. Inside this category, Terminal-Bench 2.0 is the benchmark that creates the most daylight between them.
For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.
Free. No spam. Unsubscribe anytime.