GovTools

Catalog-metadata extraction — model comparison

Which model best extracts catalog metadata (title, document type, date, subjects, named entities, issuing body) from government documents. 27 models × 13 documents, generated 2026-06-25. Companion to the Interpolation eval (reasoning over tables and charts in PDFs).

Reading this table

Click any column to sort. Accuracy = macro mean of per-field agreement vs a frontier-consensus silver reference (panel gpt-4.1 / claude-opus-4-8 / gemini-2.5-flash, grok-4 arbitrating). Rows marked * are panel members — they define the reference, so their accuracy is circular/inflated; judge them by cost, not accuracy. Completed = documents that returned a usable result / attempted. Valid schema = share of outputs that parsed into the controlled {value, code} schema (instruction-following, not correctness). $/1k docs = approximate cost per 1,000 documents (observed tokens × public list prices); locals run free. Sample = 13 docs (directional).

All models — hosted & local

01 · sortable
ModelTierAccuracyCompletedValid schemaLatencyResp tok$/1k docs
anthropic:claude-opus-4-8hosted0.93*10/108.3s689$89.37
gemini:2.5-flashhosted0.86*10/107.8s526$1.91
xai:grok-4hosted0.8513/136.9s231$9.04
openai:gpt-4.1hosted0.84*10/104.1s327$6.12
anthropic:claude-sonnet-4-6hosted0.8310/1010.6s588$14.76
mistral:largehosted0.7812/13100%28.1s483$7.02
anthropic:claude-haiku-4-5hosted0.7713/134.9s478$4.39
openai:gpt-4o-minihosted0.5713/133.0s182$0.37
ollama:granite3.3-8blocal0.5710/1014.5s277free
ollama:gemma2-9blocal0.5410/1010.7s212free
ollama:qwen2.5-7blocal0.4910/1010.0s219free
gemini:2.5-prohosted0/30.0s0
ollama:apertus-8blocal10/1017.3s314free
ollama:mistral:7blocal10/1013.6s278free
ollama:qwen2.5-coder-7blocal10/1010.4s228free
ollama:gemma-sealion-27blocal10/1069.8s348free
ollama:phi3-minilocal9/106.6s277free
ollama:qwen2.5-3blocal9/105.6s282free
ollama:olmo2:7blocal9/107.6s141free
ollama:eurollm-9blocal8/1012.6s263free
ollama:apertus-70blocal8/10357.4s307free
ollama:climategpt-70blocal8/10163.9s356free
ollama:llama3local6/107.4s159free
ollama:climategpt-13blocal6/1023.9s175free
ollama:apertus-70b-q3local4/10902.8s178free
ollama:climategpt-7blocal3/107.2s112free
ollama:deepseek-r1-8blocal1/1023.8s499free

Per-field accuracy — every model

02

Accuracy is field-dependent, so the cheapest correct pipeline routes each field to the best (non-panel) model for it rather than picking one model overall. Each cell is per-field agreement accuracy; the highlighted cell is the best non-panel model for that field (where you would route it).

Fieldopus°gemini°grok-4gpt-4.1°sonnetmistralhaiku4o-minigranitegemmaqwen
jurisdiction_code1.001.001.001.001.001.000.900.800.901.000.70
language_code1.001.001.001.001.001.001.000.900.901.000.90
document_type_code1.001.001.000.780.890.830.780.890.440.500.67
coverage_codes1.000.291.001.001.000.830.830.300.400.220.30
coverage_text0.930.801.001.000.930.900.640.640.430.580.79
issuing_body_name1.001.000.880.940.940.881.000.620.810.790.50
title0.811.000.560.810.690.830.750.310.380.440.25
subjects0.750.710.450.620.590.430.600.360.410.420.43
key_points0.810.780.650.730.740.590.720.420.520.470.36
abstract_summary1.001.001.000.500.500.500.500.500.500.000.00

° panel / reference-defining model (gpt-4.1, opus, gemini-flash) — its accuracy is circular and shown muted, not eligible as a routing target. Highlighted = best non-panel model per field. Columns ordered by macro accuracy.

The cost story

The current catalog pipeline runs on grok-4.20 (~grok-4 tier, ~0.74 accuracy here). Field-completeness alone suggested cheap/free models were equivalent — but accuracy tells a sharper story: free local models emit perfect schema yet get more values wrong (≈0.49–0.57), and gpt-4o-mini (~24× cheaper) drops to ≈0.57. The strongest genuinely-cheap, non-panel option is claude-haiku-4-5 (≈0.77).

Because accuracy is field-dependent (jurisdiction codes are easy for everyone; document type and title are not), the cheapest correct pipeline is per-field routing (table 02) rather than one model overall. Reference is frontier-consensus silver on 13 docs — directional, not a human gold set.

← Back to Model Evals overview