The changelog.
Every time the pipeline runs it appends one row: the Software & ML research (SWE) forecast, and what each of the four field-wide sources said, in that source’s own units, frozen as of that date. Nothing is smoothed. A number quoted from this site can be traced to the row it came from.
READINGS OVER TIME
four units, four axesFrontier computelog₁₀ FLOP
Capable inference$ / 1M tokens
2 of 12 runs carry no reading
Top ratingElo
Marketprobability
2026-08-1912 runs on record2026-10-06
| Date | SWE median year | SWE P10 year | SWE P90 year | SWE P by 2030 | Compute log₁₀ FLOP | Cost $/Mtok | Rating Elo | Market prob. |
|---|---|---|---|---|---|---|---|---|
| 2026-10-06 | 2027 | 2026 | 2028 | 100.0% | 26.70 | $0.090 | 1508 | 38% |
| 2026-09-07 | 2027 | 2026 | 2028 | 99.9% | 26.70 | $0.090 | 1508 | 38% |
| 2026-09-06 | 2027 | 2026 | 2028 | 99.9% | 26.70 | $0.090 | 1508 | 38% |
| 2026-09-05 | 2027 | 2026 | 2028 | 99.9% | 26.70 | $0.090 | 1508 | 38% |
| 2026-09-02 | 2027 | 2026 | 2028 | 99.9% | 26.70 | $0.090 | 1508 | 38% |
| 2026-08-30 | 2027 | 2026 | 2028 | 100.0% | 26.70 | $0.090 | 1508 | 38% |
| 2026-08-29 | 2027 | 2026 | 2029 | 99.9% | 26.70 | $0.090 | 1508 | 38% |
| 2026-08-26 | 2027 | 2026 | 2029 | 99.9% | 26.70 | $0.090 | 1508 | 38% |
| 2026-08-25 | 2027 | 2026 | 2029 | 99.9% | 26.70 | — | 1508 | 38% |
| 2026-08-24 | 2027 | 2026 | 2029 | 99.9% | 26.70 | $0.017 | 1508 | 38% |
| 2026-08-23 | 2031 | 2029 | 2034 | 31.9% | 26.70 | $0.100 | 1508 | 38% |
| 2026-08-19 | 2032 | 2025 | 2043 | 42.3% | 26.70 | — | 1384 | 67% |
Rows are written by pipeline/model/run.py. Re-running on a date that already exists replaces that row rather than adding a duplicate.