NETHERLANDSResearch
Legal Benchmarks DELTA finds no LLM ready for independent Dutch legal research
Legal Benchmarks on September 10, 2026 released its DELTA Benchmark Report scoring 23 large language models on more than 200 Dutch legal research tasks. Legaltech News reported that no model was ready for independent Dutch legal research; Claude Fable 5.1 led with a 15.0% task pass rate. Lawyers at 12 partner firms helped review tasks and criteria.
Paper (Legaltech News) → · Legal Benchmarks, Recht.nl
Key points
- Legaltech News: Legal Benchmarks announced the DELTA report on Thursday, September 10, 2026.Legaltech News
- Twenty-three models from Anthropic, Google, OpenAI and others ran more than 200 Dutch legal research tasks.Legaltech NewsLegal Benchmarks
- DELTA found no model ready to conduct Dutch legal research independently.Legal BenchmarksLegaltech News
- Claude Fable 5.1 led with a 15.0% task pass rate; Claude Opus 5 led criteria pass rate at 77.6%.Legal Benchmarks
- Grok 4.6 offered the strongest balance of legal quality and cost among scored models.Legal Benchmarks
- DeepSeek V4 Pro was the strongest open-weight model, matching Astra's 6.7% task pass rate at lower cost.Legal Benchmarks
Sources3 articles across 3 outlets
Legal Benchmarks
Legaltech News
Recht.nl