Applied Compute claims 100x lower code-search cost than a frontier model
Applied Compute says it partnered with turbopuffer to train the open-weight Qwen3.6-35B-A3B model to find code across roughly 9,000 repositories.
TLDR
Applied Compute says its 35-billion-parameter, open-weight model searches a precomputed index to answer repository-search questions at 100x lower cost than a frontier model. The company says Qwen3.6-35B-A3B, trained in partnership with turbopuffer, also tops the needle-in-a-haystack task at 2–10x lower latency.
Combined views
—
1 Source, first seen ago
