Internal · evaluation

Model evaluation lab

Candidate summary models are scored blind by CRS reviewers against CRS summary standards, revision 6 on a rotating sample of 250 measures. Models that fall below standard are not eligible for assisted drafting.

Run date

Feb 27, 2026

Measures evaluated

250

Reviewers

Pass rate with reviewer sign-off

96%

Model-only pass rate

78%

Scores by criterion

Mean reviewer score per model, 0 to 100, against CRS summary standards, revision 6.

Candidate models

Same measure, same prompt. Reviewers score without knowing the vendor.

Model	Vendor	Accuracy	Completeness	Neutrality	Plain language	Pass rate	Eligibility
Model A	ChatGPT (OpenAI)	92	88	96	90	91%	Eligible
Model B	Claude (Anthropic)	95	91	97	93	94%	Eligible
Model C	Gemini (Google)	86	79	84	88	82%	Below standard
Model D	Perplexity	90	94	92	81	89%	Eligible
Model E	Copilot (Microsoft)	88	90	89	85	87%	Eligible

Model A

91%

ChatGPT (OpenAI) · sample output for the same measure.

This bill makes permanent Medicare payment for telehealth services at rural originating sites and authorizes grants for network capacity at clinical sites.

Model B

94%

Claude (Anthropic) · sample output for the same measure.

The bill permanently authorizes telehealth payment for rural sites, sets payment parity with in-person care, and funds site connectivity through 2031.

Model C

82%

Gemini (Google) · sample output for the same measure.

This measure would expand telehealth access for rural patients and provide funding for clinic broadband upgrades over five fiscal years.

Model D

89%

Perplexity · sample output for the same measure.

The bill amends section 1834(m) of the Social Security Act, adds a voluntary interstate licensure program, and requires annual reporting for four years.

Model E

87%

Copilot (Microsoft) · sample output for the same measure.

This bill covers rural telehealth payment, grant funding limits per site, licensure coordination, reporting duties, and authorized appropriations.