The Claim LedgerWhere vendor claims outrun evidence — ranked by verification debt.

239 models carry 2467 tracked claims, 31% independently reproduced. 2 contradicted by an outside measurement; the board ranks where verification is most overdue. Every row links to the model's full ledger with sources.

Model	Claims	Reproduced	Self-rep.	Unverified	Contradicted
DeepSeek-R1	17	2	4	10	1
DeepSeek-V3	19	3	10	5	1
bloomz-560m	51	0	0	51	0
GPT-5	30	10	0	20	0
o3	27	8	0	19	0
Claude Opus 4.5	28	9	1	18	0
GPT-5.5	28	10	0	18	0
Gemini 3 Pro	25	7	0	18	0
Claude Sonnet 4	23	5	0	18	0
Gemini 3.1 Pro	28	10	1	17	0
Gemini 2.5 Pro (Jun 2025)	24	7	0	17	0
Claude 3.7 Sonnet	21	5	0	16	0
Claude Sonnet 4.5	26	10	1	15	0
Claude Opus 4.6	25	9	1	15	0
GPT-5.2	24	9	0	15	0
Claude Opus 4.8	23	8	0	15	0
o4-mini	22	7	0	15	0
Claude Opus 4	21	6	0	15	0
GPT-4o (Nov 2024)	21	5	1	15	0
GPT-5.4	25	10	1	14	0
Claude Opus 4.7	24	10	0	14	0
Grok 4	21	6	1	14	0
Claude Sonnet 4.6	21	8	0	13	0
GPT-5.1	20	7	0	13	0
Claude Fable 5	21	9	0	12	0
Gemini 3 Flash	20	8	0	12	0
Kimi K2.5	19	7	0	12	0
Claude 3.5 Sonnet (October 2024)	18	5	1	12	0
GPT-5.6 Sol	20	9	0	11	0
GLM-5.2	19	8	0	11	0
Kimi K3	18	7	0	11	0
Grok 4.5	17	6	0	11	0
o1	17	5	1	11	0
o3-mini	17	6	0	11	0
PowerMoE-3b	11	0	0	11	0
Claude Opus 4.1	19	9	0	10	0
Claude Opus 5	18	8	0	10	0
GPT-5 mini	18	8	0	10	0
gpt-oss-120b	14	4	0	10	0
Llama 4 Maverick	14	4	0	10	0
GPT-4.5	13	3	0	10	0
Gemini 2.5 Pro (Mar 2025)	12	2	0	10	0
Llama 2-13B	11	0	1	10	0
Llama 2-7B	11	0	1	10	0
Claude Haiku 4.5	16	7	0	9	0
Claude Sonnet 5	16	7	0	9	0
GPT-4.1	16	7	0	9	0
GPT-5.6 Luna	16	7	0	9	0
GPT-5.6 Terra	16	7	0	9	0
Grok 3	14	5	0	9	0
DeepSeek-R1 (May 2025)	13	4	0	9	0
Gemini 3.5 Flash	17	9	0	8	0
Grok 4.6	16	8	0	8	0
GPT-4o mini	15	4	3	8	0
DeepSeek-V3.2	14	6	0	8	0
GPT-5.4 Mini	14	6	0	8	0
Claude 3.5 Haiku	13	5	0	8	0
Gemini 1.5 Pro (Sept 2024)	11	3	0	8	0
LLaMA-13B	11	0	3	8	0
LLaMA-7B	11	0	3	8	0

Showing the 60 highest-verification-debt models of 239 with tracked claims. Statuses are defined in the methodology; each model page carries the full ledger with sources.