Vals, Keenable
Vals, Keenable
Keenable ranks #1 on the Vals Web Search Index
Keenable ranks #1 on the Vals Web Search Index
Keenable ranks #1 on the Vals Web Search Index
Keenable Pro gives GPT-5.6 Sol the highest overall score, at 53% lower cost per task than the same model with Exa.
Vals AI has added Keenable Pro to its Web Search Index. Paired with GPT-5.6 Sol, Keenable ranks first overall at 49.08% across finance analysis and legal research.
The benchmark holds the model and agent harness constant while changing the search tool, measuring how that choice affects the agent’s final answer.
Keenable Pro gives GPT-5.6 Sol the highest overall score, at 53% lower cost per task than the same model with Exa.
Vals AI has added Keenable Pro to its Web Search Index. Paired with GPT-5.6 Sol, Keenable ranks first overall at 49.08% across finance analysis and legal research.
The benchmark holds the model and agent harness constant while changing the search tool, measuring how that choice affects the agent’s final answer.
Higher scores with the same model
With GPT-5.6 Sol, Keenable Pro scores 5.50 percentage points above native search and 3.84 points above Exa.
Those gains appear in both domains. GPT-5.6 Sol reaches 53.9% in finance and 44.2% in legal with Keenable, compared with 52.0% and 38.5% with Exa.
53% lower cost per task than Exa
With GPT-5.6 Sol, Keenable’s average cost per task is $8.84, compared with $18.97 using Exa. That is 53% lower total task cost, including both model inference and search fees.
Evaluated on professional tasks
The index combines 450 finance-analysis questions from FAB v2 with 208 expert-authored legal-research tasks. General web search is the agent’s only source of external information; domain-specific tools such as SEC EDGAR and CourtListener are removed.
Legal answers must satisfy every requirement in an expert-written rubric to pass. Finance answers earn partial credit, but failing a critical fact or number check reduces their score to zero.
Read the Vals methodology for the full evaluation setup and scoring rules.
Get started with Keenable
Keenable is available through our Search API and MCP. Run it in your own agent and compare answer quality, total task cost, and completion time on the tasks your users care about.
Documentation | Create an account
Results from Vals AI’s October 19, 2026 evaluation. Keenable has been evaluated with GPT-5.6 Sol.
Higher scores with the same model
With GPT-5.6 Sol, Keenable Pro scores 5.50 percentage points above native search and 3.84 points above Exa.
Those gains appear in both domains. GPT-5.6 Sol reaches 53.9% in finance and 44.2% in legal with Keenable, compared with 52.0% and 38.5% with Exa.
53% lower cost per task than Exa
With GPT-5.6 Sol, Keenable’s average cost per task is $8.84, compared with $18.97 using Exa. That is 53% lower total task cost, including both model inference and search fees.
Evaluated on professional tasks
The index combines 450 finance-analysis questions from FAB v2 with 208 expert-authored legal-research tasks. General web search is the agent’s only source of external information; domain-specific tools such as SEC EDGAR and CourtListener are removed.
Legal answers must satisfy every requirement in an expert-written rubric to pass. Finance answers earn partial credit, but failing a critical fact or number check reduces their score to zero.
Read the Vals methodology for the full evaluation setup and scoring rules.
Get started with Keenable
Keenable is available through our Search API and MCP. Run it in your own agent and compare answer quality, total task cost, and completion time on the tasks your users care about.
Documentation | Create an account
Results from Vals AI’s October 19, 2026 evaluation. Keenable has been evaluated with GPT-5.6 Sol.
