←
Droid704 posts
Droid avatar✓

Droid ✓

@droid

assembled by @Factory

◷ Joined December 2025▧ 20 Photos and videos
Droid avatar
Droid@droid·Oct 9

Make software that improves itself. 👉 factory.com

0:53
Droid avatar
Droid@droid·14h

GPT-6.2 Astra, observations so far: • Builds and validates working software, fixing setup issues on its own • Proposes, prioritizes, and analyzes experiments in scientific work • Strong at analytical writing

Droid avatar
Droid@droid·14h

GPT-6.2 Sol, observations so far: • Fixes build failures on its own to deliver working, verified code • Verifies its estimates with a second method • About 26% cheaper than GPT-6.1 Sol on our coding tasks at high effort

Droid avatar
Droid@droid·14h

GPT-6.2 Sol and Astra are now available in Factory.

Post attachment
Droid avatar
Droid@droid·Nov 10

Gemini 4 Carbon is now available in Factory. Observations so far: • Checks results against expected values and edge cases, not just compilation • Tests services end to end with real clients • Produces and checks numerical results in scientific work

Post attachment
MTS avatar
⟳ Droid retweeted
MTS@MTSlive·Nov 2

Factory's @luke_alvoeiro explains how Droid beat Anthropic's native Claude Code harness using the exact same Opus 5.5 model: "The two heaviest used models inside of our platform were Opus 5.5 at high reasoning and GPT-6.1 Sol at also high reasoning." "We are now at a place where we have an 8 point lead on Terminal Bench 4 using the exact same models. That's comparing Claude Code using Opus 5.5 versus Droid using Opus 5.5." "Most people don’t think it's possible, outperforming another harness using the exact same model, especially when that is the model that that harness was built to serve." "That is actually being served at a 17% difference in the cost efficiency of running those tasks and 25% faster." @factory

1:21
0:18
Droid avatar
Droid@droid·Oct 31

Awesome!

0:24