TrueFoundry TrueForge cuts AI costs by 75% over Claude

TrueFoundry, a San Francisco‑based machine‑learning startup, today announced the release of TrueForge, an open‑source harness for AI agents. The tool is licensed under the MIT licence and can be paired with any large‑language model (LLM). In tests, TrueForge completed 11 of 14 tasks on the DevRev Enterprise‑Bench at a cost 75 % lower than Anthropic’s Claude Managed Agents, paying only $2.90 for the same workload.

The launch comes amid a surge of AI agent deployments in business software. Companies want agents that can use tools, access internal data, and stay under budget. TrueFoundry, founded in 2021 by former Meta engineers, has built its product to give developers full control over the agent stack and to keep costs predictable. By making the harness open source, the company removes the lock‑in that many commercial agents impose.

In the benchmark, TrueForge was run with the open‑source GLM‑5.2 LLM. It handled multi‑step tool use across customer‑relationship management, issue tracking and document‑management systems. The harness achieved 11 out of 14 successful completions, while Claude Managed Agents required a higher‑tier model and incurred a 75 % higher price tag. The $2.90 figure reflects the cost of running the GLM‑5.2 model on the same hardware, compared with the $10.30 that Claude’s Opus 4.8 would have cost for the same task set.

The move could reshape how enterprises build AI‑powered workflows. With TrueForge, developers can fork the code, modify it, host it on their own servers, and embed it in commercial products without licensing fees. As the market looks for cheaper, more flexible AI solutions, TrueForge offers a compelling alternative that keeps the power in the hands of the business rather than the vendor.