Run 1
4:59 min:stime
12,017output tokens
1.9%thinking
40.17tok/s
666lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.5-35b-a3b_bare_r1
Ornith 1.5 35B A3B puts 95+ effects on every page. Do rules cut them?
ornith-1.5-35b-a3b
Same model, same prompt, same rig. 3 runs with nothing added, 3 with one Karpathy rules file.
ornith-1.5-35b-a3b
Ranges overlap (counted effects)Effects count CSS/JS constructs, not visual quality.
Repeated ranges
100–117
KARPATHY · n=395–117
3:56–4:59 min:s
KARPATHY4:48–6:14 min:s
10,079–12,017
KARPATHY11,637–13,902
40.17–42.65 tok/s
KARPATHY37.18–40.44 tok/s
0.8%–1.9%
KARPATHY2.4%–2.6%
609–719
KARPATHY646–763
BARENO RULES, NO EXTRAS
4:30 min:stime
11,140output tokens
0.8%thinking
609lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
KARPATHYONE RULES FILE
5:03 min:stime
12,094output tokens
2.5%thinking
681lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
Every run. Every result.
BARE
NO RULES, NO EXTRAS · n=3
Run 2
4:30 min:stime
11,140output tokens
0.8%thinking
41.21tok/s
609lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.5-35b-a3b_bare_r2
Run 3
3:56 min:stime
10,079output tokens
1.2%thinking
42.65tok/s
719lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.5-35b-a3b_bare_r3
KARPATHY
ONE RULES FILE · n=3Run 1
95effects
4:48 min:stime
11,637output tokens
2.4%thinking
40.44tok/s
763lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.5-35b-a3b_karpathy_r1
Run 2
117effects
6:14 min:stime
13,902output tokens
2.6%thinking
37.18tok/s
646lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.5-35b-a3b_karpathy_r2
Run 3
116effects
5:03 min:stime
12,094output tokens
2.5%thinking
39.92tok/s
681lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.5-35b-a3b_karpathy_r3
output = thinking + final code. Lines of code show the page size the model chose, not the cost of equivalent work.
How to reproduce
Settings recorded for these runs. Multiple values show differences between runs. Effort is requested, not necessarily received; it is not comparable across models. Hover over quantization, K/V cache or engine values for their sources.
ornith-1.5-35b-a3b
| Model | Quantization | Context window | K/V cache | max_tokens | Engine | Harness | Effort requested |
|---|---|---|---|---|---|---|---|
| ornith-1.5-35b-a3b | Q4_K_M | 262,144 | K q4_0 / V q4_0 | 131,072 | [email protected] | pi 0.85.1 | medium |
Watch the experiment.
REPRODUCE THE EXPERIMENT
One prompt.
All the evidence.
Compare runs within the same model. Inspect the shared prompt, settings and raw measurements in the repository.