Atlas

Models

← All models

Ember-1

Fireworks AI · 2026-09-23

Fireworks' hosted reasoning model post-trained from Kimi K3 to produce shorter reasoning traces for coding, knowledge work, and agentic tasks. Fireworks announced the model on September 23, 2026; its hosted profile supports image input and function calling.

Benchmark scores

BenchmarkScore
Bedside Bench89.1
BigFinanceBench private holdout (final-answer accuracy)39.0
BioMysteryBench — Hard (Vals)29.4
BioMysteryBench — Human-solvable (Vals)75.8
BioMysteryBench (Vals Terminus 2)67.0
BuseyBench SVG5.8
Capability160.5
Code Migration32.5
Code Migration - CLI23.3
Code Migration - COBOL60.0
Code Migration — Code Quality4.0
Code Migration (Vals Index v2 subset)30.5
CyberBench v1.1 Overall69.5
CyberBench v1.1 Patch85.7
CyberBench v1.1 PoC53.3
CyberGym (pass@3)89.7
DeepSWE v1.175.2
DeepSWE v1.1 (Fireworks, avg@3)67.0
Excel Modeling Benchmark63.4
Excel Modeling Benchmark Comps60.5
Excel Modeling Benchmark Dataroom81.2
Excel Modeling Benchmark DCF50.3
Excel Modeling Benchmark LBO58.1
Excel Modeling Benchmark M&A66.7
Excel Modeling Benchmark Misc63.1
Excel Modeling Benchmark Operating63.7
Excel Modeling Benchmark Scratch65.6
Excel Modeling Benchmark Template60.8
Finance Agent v251.9
Finance Agent v2 - Adjustments51.3
Loading Atlas data…