Same task. Same pass. Half the wall time.

Two coding agents refactor the same Momento repo task live. Opus 4.8 and GLM-5.2 both pass the tests. GLM-5.2 on Baseten finishes 2.1x faster, costs 2.9x less, and emits far fewer output tokens.

Watch the race See the scorecard

Opus 4.8 19.5 min $4.65 with 89k output tokens

GLM-5.2 on Baseten 9.3 min $1.60 with 27k output tokens

Correctness PASS Both agents complete the task and pass validation.

Efficiency 2.9x Cheaper on the same passing outcome.

Live coding-agent comparison

The race, with the finish held long enough to read.

The animation pauses on the final frame for 10 seconds so the takeaway is legible: both pass, but GLM-5.2 reaches the finish with dramatically less time and spend.

Opus 4.8 · Claude Code vs GLM-5.2 Baseten · mo final frame +10s

1. Same task, side by side

Both agents work on the same refactor task in the same repo, making the comparison easy to follow.

2. GLM moves with less output

The GLM-5.2 run produces materially fewer output tokens, which lowers cost and reduces review noise.

3. Both reach PASS

The important quality bar is not speed alone. Both runs pass the task validation.

4. The finish line is economic

For a passing result, GLM-5.2 finishes 2.1x faster and 2.9x cheaper.

Result summary

Passing code at a better operating point.

The high-level story is simple: the task passes on both models, but GLM-5.2 on Baseten uses less time, less money, and fewer generated tokens.

Opus 4.8 · Claude Code

GLM-5.2 Baseten · mo

Result

PASS

Cost

$4.65

$1.60 (2.9x cheaper)

Wall time

19.5 min

9.3 min (2.1x faster)

Tokens out

89k

27k

For agentic coding, the backend should make the same pass cheaper and faster.

Maude can route coding-agent work through the model lane that fits the job, while keeping the workflow familiar for developers and measurable for the team.

PASS  both agents complete the task
2.1x  faster wall-clock finish
2.9x  cheaper total run
27k   output tokens instead of 89k