SKILL.state: Input, Cache, and Output
This note follows the first article on the SKILL.state experiments, which describes the protocols, run conditions, and criteria for full success.
The data in this note come from the Codex series run on September 8–9: two models, four modes, five projects, and five repetitions of each project. Each row covers 25 sessions; the totals include timeouts and failed runs.
Input counts tokens in the main agent loop, including cached tokens. “Cached share” is the proportion of these tokens that the provider classified as cached input. Output counts tokens generated by the model in the main agent loop.
| Model | Mode | Input, M tokens | Cached share | Uncached input, M tokens | Output, K tokens | Full success |
|---|---|---|---|---|---|---|
| Sol | Native | 6.367 | 89.4% | 0.673 | 270.8 | 24/25 |
| Sol | Paper | 9.215 | 78.9% | 1.946 | 402.7 | 19/25 |
| Sol | V2 | 4.077 | 52.3% | 1.943 | 221.4 | 23/25 |
| Sol | V3 | 4.180 | 47.2% | 2.207 | 251.9 | 25/25 |
| Astra | Native | 2.734 | 85.3% | 0.402 | 150.0 | 25/25 |
| Astra | Paper | 15.318 | 75.0% | 3.833 | 324.5 | 16/25 |
| Astra | V2 | 2.654 | 63.7% | 0.964 | 133.6 | 25/25 |
| Astra | V3 | 2.340 | 56.5% | 1.018 | 130.5 | 25/25 |
Modes: Native — the standard agent loop; Paper — the first adaptation of SKILL.state with a single text observation; V2 — structured state and a three-observation window; V3 — V2 with batches of actions executed sequentially.
What the numbers show
V2 and V3 reduce total input and output relative to Native for both models. For Sol V2: input −36.0%, output −18.2%. For Astra V3: input −14.4%, output −13.0%.
The reduction in total input comes with a smaller share of cached tokens. For Sol, that share falls from 89.4% in Native to 52.3% in V2; for Astra, from 85.3% to 56.5% in V3. Uncached input increases in both cases.
Paper increases both input and output token volumes and completes tasks less often. For Astra, it used 5.6 times as much input and 2.2 times as much output as Native; full success was 16/25 versus 25/25.
For Sol, V2 uses less input and output than V3, but achieves full success in 23/25 runs versus 25/25. For Astra, V3 uses less input and output than V2, with the same full success rate.
The cached share describes the composition of input tokens, not the proportion
of requests that hit the cache. It does not equal financial savings: calculating
cost requires separate rates for uncached input, cached input, and output.
Output also does not measure the quality or amount of the final code.