Fireworks Ember-1 Release - SWE-Interact: leaderboard

Fireworks Research Ember-1 launch post (2026-09-23): controlled four-arm evaluation on the 75-task SWE-Interact multi-turn software-engineering suite comparing Ember-1 (built on Kimi K3; reasoning effort unreported/default) against Fireworks' runs of Kimi K3 at reasoning efforts low, high and max (default). Agent harness, scaffold and repeat count are not disclosed in the post. Pass Rate (%), higher is better.

Metric: Pass Rate (%). Source: fireworks.ai. Saturation forecast: Around 2028. 2 models tracked. Updated 19 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Kimi K3Max21.3#61
2Ember-120#92

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=fireworks-ember-1-release-swe-interact · How It Works · Data refreshed daily, snapshot 2026-11-20.