We just launched “Enterprise-Bench” – the AI industry’s first open, vendor-neutral benchmark built specifically for enterprise use cases.
And we’re making a challenge to other AIs: come test yourself against Computer by DevRev.
AI companies throw around lots of benchmarks – but they mostly test single-user tasks that don’t run inside the organizational complexity of a real enterprise.
We built Enterprise-Bench to fill that gap. We ran Computer against Claude Code. Same Opus 4.8 model. Same dataset. Same independent judge.
The only variable was architecture.
– ACCURACY. Computer: 94.3% vs. Claude: 63.6%.
– TOKEN EFFICIENCY. Computer: 4.4x fewer tokens per correct response.
– COST AT SCALE.
As the dataset grew, Computer’s token usage: stayed roughly flat vs. Claude Code: climbed by 29%.
Independently validated by Professor Alexandros Dimakis (University of California, Berkeley).
Built on Terminal Bench. Full dataset on Harbor Hub. Full GitHub access.
The leaderboard is open.
Any vendor, researcher, or enterprise can take the same test and submit their results: dvrv.ai/4pvkwZX
So, who’s going to step up?
12 Followers 97 FollowingProof layer for AI systems. Independently verifiable evidence for AI decisions, without storing raw prompts or outputs. Private beta.
2K Followers 695 FollowingGraph technology specialist. Yes, that includes Semantic Technologies too; PhD in AI/KR; Co-host of https://t.co/TRLtxCH4JF
I work for @Neo4j.
19K Followers 947 FollowingCEO @DevRev. Board Member @Adobe. Constant Student of Business/Tech. Beachcomber of People Experiences. Worshipper of Authenticity, Hunger & Design