←Story detail

Trending story

Fireworks AI launches the Specialized Intelligence Index

5 reports5 story updatesUpdated Sep 23 08:00

Understand this story

AI overview

On September 22, 2026, Fireworks AI released the Specialized Intelligence Index (SII), which brings together the performance of open, closed-source, and specialized models on domain-specific benchmarks, initially covering seven domains: healthcare, legal, cybersecurity, finance, customer service, productivity, and software. According to the official description, each benchmark was designed by practitioners in the field, such as Harvey's Legal Agent Benchmark, Sierra's τ-Banking and τ-Voice, and Datacurve's DeepSWE v1.1. SII does not weight or average results across domains; it retains only benchmark- and domain-level scores and provides comparison views of score versus cost and score versus duration. Each benchmark is annotated with its reproducible scope, and new frontier models must pass a production deployability screening before being included. Fireworks also accepts partner submissions, with partners providing tasks and data in a Harbor-compatible format; Fireworks runs the evaluations free of charge and publishes scores with the partner's consent. There have been no further developments so far.

AI-generated from public reports

Latest development · Report

Fireworks Research Releases Ember-1: A Token-Saving Specialized Model Based on Kimi K3

Developments

1 confirmed developments

  1. Fireworks AI launches Specialized Intelligence Index

    Source:Fireworks AI

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Sep 23

  1. Fireworks AIFirst-party

    Fireworks Research Releases Ember-1: A Token-Saving Specialized Model Based on Kimi K3

    Fireworks Research has released Ember-1, a specialized model trained on Kimi K3 that reduces token consumption by about 40% while maintaining equivalent quality. The company says that across 7 benchmarks and production traffic from two customers, K3's reasoning process can be shortened by 35–50% without any drop in accuracy; on Bedside Bench, Ember-1 sits on the Pareto frontier for cost and task completion. The model is available on Serverless as a Research Preview, offered alongside the base Kimi K3, with training support on enterprises' own data also opened up.

  2. fireworks_aiFirst-party

    English translation is being prepared.

Sep 22

  1. fireworks_aiFirst-party

    English translation is being prepared.

  2. fireworks_aiFirst-party

    English translation is being prepared.

  3. fireworks_aiFirst-party

    English translation is being prepared.

Related stories

  1. Related storiesMeta launches Muse Wonder and Muse Wonder Full models
  2. Related storiesNvidia releases Nemotron 3.5 Super VL and Bolt models
  3. Related storiesMeta releases open-weight model Muse Spark Open
  4. Related storiesEnglish translation is being prepared.