Keenable.ai ads

If you run Hermes, your agent has already been searching on Keenable. Last month traffic grew 8x, and we enabled 1M search requests a day from Hermes agents. Hermes agent’s ecosystem keeps growing, add your own key to lift the default limits.

Keenable Pro ranks #1 on the Vals Web Search Index with GPT-5.6 Sol. Results across finance and legal research tasks: #1 overall out of 9 configurations, with 49.08% accuracy #1 in finance analysis, with 53.9% accuracy 53% lower total cost per task: $8.84 vs. $18.97 26% shorter task duration: 6m 35s vs. 8m 51s Cost and duration are compared with Exa using the same model, agent harness and tasks. Total cost includes model tokens and search fees.

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.

Keenable Search is now native to Baseten. You turn on hosted web search and page fetch in your Baseten request, Baseten runs the tool and passes the retrieved content back to the model. The model searches, reads, and keeps reasoning inside that same request. No separate search API key and agent loop to manage now. We want models to search more: to check an assumption instead of guessing, and to follow a lead before answering. Baseten is our first inference-platform partner, and context management, quality, and latency are what we'll keep working on.

You can’t build a self-learning system without a reliable way to measure it. We found that AI is great in finding loopholes in static benchmarks. This breaks the learning loop. Check out our post co-authored by Ilya Gusev and Matthias Petri on what is broken in BrowserComp and other static evals. Today we introduce NEEDLE, a live, open-source web search benchmark designed to address these issues.

Agents still search through ten blue links. We wrote down what they could get instead. With Keenable SELECT, agents can get structured data from the internet on the fly. In the article, we explain when snippets are not enough and show how agents can use Keenable SELECT to retrieve distributions of data points and reason over them—not just individual search results. It works through SQL-like queries and gives agents structured dataframes they can compute over directly. Search shouldn’t just return links or snippets. It should give agents data they can compute and reason over.