StepFun open-sources Step Code: 80.9% on Terminal-Bench 2.1 with fewer tokens
StepFun_ai · x · 2026-09-23
StepFun released Step Code v0.1.0, an MIT-licensed open-source coding CLI covering the full dev loop from reading/editing code to running tests and shipping.
- Terminal-Bench 2.1: 72/89 tasks passed (80.9%), tied for highest pass rate while using fewer tokens than the other tied leaders
- Multi-Frame, its own 150-task long-horizon benchmark: 110/150 passed (73.3%), averaging 5.09M tokens/task — highest pass rate and lowest token use among six harnesses
- Ships with StepPage: one command publishes a local static site to a shareable URL, with versioning and rollback built in
More from coding & agent
- Hamel Husain: refresh stale eval datasets with regular error analysis, retire passing ones — HamelHusain · 2026-09-23
- Claude Rebuilds the Iconic Earthrise in 3D, Down to the Second — Simple-Diver-2192 · 2026-09-23
- Deel automates 85% of payment reconciliation by having an agent watch its best operator work — aryanXmahajan · 2026-09-23
- Deel's Akai agent platform claims $140M ARR in 90 days with record-once workflows — aryanXmahajan · 2026-09-23
- Hexis Shares How They Manage Agent Skills Permissions on Top of Git — juanviera23 · 2026-09-23
- Claude Code 2.1.280 ships 114 changes, makes Opus 5.5 with 1M context the default — ClaudeCodeLog · 2026-09-23