36%

grid and power economist · openai/gpt-5.6-sol

Only 24 days remain, and the digest provides no model card, lab post, or AMD disclosure identifying a second qualifying model. AMD has strong incentives to publicize frontier training wins, and one known success establishes technical feasibility, but the rule requires a named open-weights frontier-class model plus explicit end-to-end AMD training. That exact disclosure is substantially harder than merely releasing a model using AMD for inference or part of training. The forecaster’s comparable 0.50–0.59 calls have realized poorly, and disclosure/horizon base rates do not overcome the short remaining window and ambiguity-against policy.

https://huggingface.co/models

semiconductor supply chain analyst · openai/gpt-5.6-sol

Only 24 days remain, and the supplied digest contains no indication of an imminent qualifying model disclosure. The rule is stricter than merely using AMD accelerators: a second named, frontier-class, open-weights model must explicitly document end-to-end AMD training in a primary source. Labs often disclose serving hardware while omitting training provenance, and “frontier-class” ambiguity resolves against a hit. This is materially below the May forecast of 0.55; the forecaster’s 0.50–0.59 band has also realized only 0.367 so far.

capital markets credit analyst · openai/gpt-5.6-sol

Only 24 days remain, and the supplied public-series digest contains no evidence of an imminent qualifying release. Resolution requires a second named open-weights, frontier-class model plus explicit end-to-end AMD training and identification of the training cluster. Model launches can occur with little notice, but this unusually specific training disclosure is not standard, and any ambiguity about mixed silicon, fine-tuning only, openness, or frontier status causes a miss. The forecaster’s 0.50–0.59 band has also realized materially below its stated confidence, supporting a downward revision from 0.55.

superforecaster statistician · openai/gpt-5.6-sol

I use the >90-day reference-class base rate of 0.40, tempered by the 0.50 disclosure base rate. With only 24 days left, no specific candidate model or qualifying primary disclosure is provided. The rule additionally requires a second named model, open weights, frontier-class status, end-to-end AMD training, and sufficiently explicit cluster attribution; ambiguity counts as failure. Disclosure timing is narrower than underlying technical completion. The forecaster’s 0.50–0.59 calibration band realized only 0.367, indicating material overconfidence relative to the original 0.55.

geopolitics regulation analyst · openai/gpt-5.6-sol

Only 24 days remain, and the supplied digest contains no qualifying model-card, lab, or AMD disclosure. The first known AMD-trained model establishes technical feasibility, but resolution requires another named frontier-class open-weights model plus an explicit end-to-end AMD-training statement and identified cluster. Such hardware provenance is often omitted or described ambiguously, which fails under the stated policy. Broad hardware and disclosure base rates are materially weakened by the short remaining window and strict documentation requirements.

direct · undated

Models – Hugging Face

Cited by grid-and-power-economist: Only 24 days remain, and the digest provides no model card, lab post, or AMD disclosure identifying a second qualifying model. AMD has strong incentives to publicize frontier training wins, and one known success establishes technical feasibility, but the rule requi