Microsoft Research AI Frontiers releases Fara1.5 computer use agent models in 4B, 9B, and 27B sizes scaling to 72% Online-Mind2Web success
27B version tops Gemini 2.5 CU and Operator on benchmarks.
@aravindr93 Neat!
You likely know this, but things get better with code generation actions:
𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐢𝐧𝐠 𝐍𝐚𝐯𝐢𝐠𝐚𝐭𝐨𝐫 𝐧𝟏.𝟓 The most capable computer-use model for the web. Pareto-domination: accuracy, latency, cost • SoTA across all benchmarks • +5-10% over GPT 5.5, Opus 4.7, n1 • +25% over Gemini • 2x faster, significantly cheaper Expanded action space • UI actions (like n1) + JavaScript generation & execution
@DhruvBatra_ Thanks @DhruvBatra_ !
Yes ofc, your blogs are a joy to read as always 😃
I agree! Things that I think matter the most for web agents: synthetic envs; code or macro actions in general; and Dagger/on-policy distillation.
Funnily, the same things we know and love in EAI!
@aravindr93 Neat! You likely know this, but things get better with code generation actions:
@DhruvBatra_ Also, if you have the bandwidth, would love to know the automated eval scores for n1.5
The leaderboard is taking submissions again I believe.
@aravindr93 Neat! You likely know this, but things get better with code generation actions: