{"passport":{"slug":"bdh","display_name":"BDH","model_type":"language","parameter_count":null,"context_window":null,"claimed_developer":"Pathway","confirmed_developer":"Pathway","developer_display":"Pathway","developer_confirmed_at":"2025-10-01T00:00:00+00:00","developer_source":"Confirmed via https://pathway.com/ (attribution date basis: press_coverage)","first_appeared_at":"2026-08-09T18:31:10.759856+00:00","first_appeared_on":"reddit_localllama","status":"confirmed","availability":null,"deployment_status":"not_deployed","availability_scope":null,"gate":3,"eu_ai_act":"GPAI","regulatory_status":{"eu_ai_act":"GPAI"},"open_source":null,"weights_available":null,"license":null,"provenance_chain":[],"parent_model":null,"model_family":null,"superseded_by":null,"supersedes":[],"recommended_replacement":null,"product_links":null},"events":[{"event_type":"developer_corrected","event_date":"2026-08-26T00:00:00+00:00","title":"Correction: disposition authority vs execution actor","detail":"The confirmation event on this row at 2026-08-26T13:32:44.197421+00:00 recorded data.disposition=\"human\" with actor=\"stark-raaw\". That is imprecise and is corrected here rather than rewritten. AUTHORITY was human — Do Raaw — by explicit ruling (Do GO x4 (S248, ~17:0x CEST), GO 2 — 'the actor each event carries; correction event per row, never a rewrite'). EXECUTION was performed by stark-raaw, an AGENT, not a person. The developer attribution and the evidence class on the original event are UNCHANGED and remain correct; only the claim about who performed the disposition is narrowed. Raised by Stark against his own record: asserting human-in-the-loop where the actor was a process is the same defect filed the same day against scripts/ack-redalert.cjs.","source_url":null,"source_name":"stark-raaw"},{"event_type":"first_appearance","event_date":"2026-08-09T18:31:10.759856+00:00","title":"First recorded in AI news","detail":"Pathway has released BDH, a post-transformer architecture model that matches GPT-2 scaling performance across parameter ranges from 10M to 1B parameters when trained from scratch. The model is designed to run on standard consumer GPUs, making it accessible for local deployment.","source_url":"https://reddit.com/r/LocalLLaMA/comments/1vjwqpf/pathways_bdhposttransformer_arch_matches_gpt2/","source_name":"reddit_localllama"}],"delta":{"days_unattributed":null,"claimed_developer":"Pathway","confirmed_developer":"Pathway","match":true}}