Myosotis-1,
measured.

INT index 6.71 release final benchmarks 3 / 4 ArithMark-3

Small language models, ranked.

The field below lists transformer-based small language models from the Intelligence Index leaderboard, with the top three, the models immediately around Myosotis-1, and Myosotis-1 itself highlighted for comparison.

Intelligence Index — small LM landscape 145 entries · sorted by index ↓
Rank Model Params Int index HellaSwag (%) ARC-easy (%) ARC-challenge (%) PIQA (%) ArithMark-3 (%)
1 SmolLM2-135MHuggingFace · 135M 135M 27.13 43.22 58.63 29.69 68.44 39.20
2 SmolLM-135MHuggingFace · 135M 135M 25.74 42.70 56.31 29.01 68.28 36.80
3 GPT-X2.5-135MAxiomic Labs · 135M 135M 25.17 40.57 51.81 29.18 69.42 38.40
⋮ · 85 models omitted (ranks 4–88) · ⋮
89 GPT-S-5MAxiomic Labs · 5.2M 5.2M 6.90 27.46 33.21 21.16 57.24 30.20
90 tinyLM-8M-expUser01110 · 8.1M 8.1M 6.82 27.65 35.65 23.61 56.51 27.50
91 Myosotis-1-base◈ our first modelFWKV · recurrent 101m 6.82 26.21 34.34 21.59 55.33 35.00
92 Atom 3.4MUCR · 3.4M 3.4M 6.37 27.69 33.33 22.10 55.17 31.60
93 Pythia-14MEleutherAI · 14M 14M 6.25 26.20 32.28 20.99 55.88 33.40
⋮ · 52 models omitted (ranks 94–145) · ⋮

Percentages are raw scores on each benchmark. Myosotis-1 uses acc_norm for HellaSwag and ARC and acc for PIQA, matching the leaderboard convention. ArithMark-3 was not evaluated, so its cell and its weight are dropped from the index.

How the Intelligence Index is calculated

Each benchmark is first adjusted for its random-chance floor, so chance performance maps to 0 and perfect performance maps to 100.

N(score, chance) = 100 × (score − chance) ÷ (100 − chance) Int Index = (HellaSwag + ARC + PIQA + 0.65 × ArithMark-3) ÷ 3.65
Benchmark Weight Chance Notes
HellaSwag 1.00 25 acc_norm
Combined ARC 1.00 25 mean of ARC-Easy + ARC-Challenge, before normalization
PIQA 1.00 50 acc
ArithMark-3 0.65 25 not evaluated for Myosotis-1 → shown as “—”
  • Combined ARC is the mean of ARC-Easy and ARC-Challenge before normalization.
  • When a component is unavailable, its weight is removed from both the numerator and the denominator; at least two components are required.
  • Scores below chance may be negative.
  • Myosotis-1 was not evaluated on ArithMark-3, so its 0.65 weight is dropped: (HellaSwag + ARC + PIQA) ÷ 3 = 6.71.
  • Myosotis-1 is a recurrent FWKV model, shown on a leaderboard of transformer-based models for comparison.