Run 1
4:19 min:stime
12,464output tokens
2.2%thinking
48.12tok/s
538lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.0-35b_bare_r1
Ornith 1.0 35B writes 945 lines for one web page. Do rules trim it?
ornith-1.0-35b
Same model, same prompt, same rig. 3 runs with nothing added, 3 with one Karpathy rules file.
ornith-1.0-35b
Ranges overlap (counted effects)Effects count CSS/JS constructs, not visual quality.
Repeated ranges
65–91
KARPATHY · n=363–98
3:09–4:19 min:s
KARPATHY3:59–4:27 min:s
9,958–12,464
KARPATHY11,528–12,266
48.12–52.73 tok/s
KARPATHY45.95–48.20 tok/s
2.2%–6.4%
KARPATHY2.5%–3.9%
538–945
KARPATHY493–835
BARENO RULES, NO EXTRAS
4:10 min:stime
12,177output tokens
5.9%thinking
945lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
KARPATHYONE RULES FILE
4:10 min:stime
12,064output tokens
2.5%thinking
709lines 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
4:10 min:stime
12,177output tokens
5.9%thinking
48.64tok/s
945lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.0-35b_bare_r2
Run 3
3:09 min:stime
9,958output tokens
6.4%thinking
52.73tok/s
864lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.0-35b_bare_r3
KARPATHY
ONE RULES FILE · n=3Run 1
76effects
4:27 min:stime
12,266output tokens
3.5%thinking
45.95tok/s
835lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.0-35b_karpathy_r1
Run 2
63effects
4:10 min:stime
12,064output tokens
2.5%thinking
48.2tok/s
709lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.0-35b_karpathy_r2
Run 3
98effects
3:59 min:stime
11,528output tokens
3.9%thinking
48.16tok/s
493lines of code
Effort requested: medium; received: on · differs from requested effort · source: metrics.json. Effort is not comparable across models.
ornith-1.0-35b_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-1.0-35b
| Model | Quantization | Context window | K/V cache | max_tokens | Engine | Harness | Effort requested |
|---|---|---|---|---|---|---|---|
| ornith-1.0-35b | 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.