Samaya AI's Sterling scores 58.9% on FrontierFinance at $0.61 per question, ahead of Claude Fable 5.1
Samaya AI has released Sterling, a finance research agent model post-trained from Nemotron 3.5 Lightning with 3B active parameters. It scores 58.9% on FrontierFinance at $0.61 per question, against 55.9% at $6.89 for Claude Fable 5.1.
Samaya AI has released Sterling, a finance research agent model that scores 58.9% on the FrontierFinance benchmark at $0.61 per question. Claude Fable 5.1 scores 55.9% on the same benchmark at $6.89 per question, according to a report from MarkTechPost.
What was released
Sterling is a post-trained version of Nemotron 3.5 Lightning, built for financial research. It is described as a 3B-active model, meaning it uses 3 billion active parameters. The report's headline puts the cost advantage over Claude Fable 5.1 at 11x.
Reported results
| Model | FrontierFinance score | Cost per question |
|---|---|---|
| Samaya AI Sterling | 58.9% | $0.61 |
| Claude Fable 5.1 | 55.9% | $6.89 |
These figures are as reported by MarkTechPost. Context window, per-token pricing, training cutoff and total parameter count have not been published in the report.
What this means
The reported comparison is on two axes: benchmark score and cost per question. On FrontierFinance, Sterling is listed with the higher score and the lower cost per question than Claude Fable 5.1. The result also shows a model post-trained from Nemotron 3.5 Lightning, with 3B active parameters, being positioned against Claude Fable 5.1 on a domain-specific research benchmark. Teams evaluating finance research agents have cost-per-question and benchmark figures to test against their own workloads.
Related Articles
Diffusion Intelligence releases Clover-35B-A3B-Preview, a 35B-parameter diffusion language model for agents
Diffusion Intelligence (扩散智能) has released Clover-35B-A3B-Preview, a mixture-of-experts diffusion language model with 35B total and about 3B active parameters per token. The team reports 73.6 on SWE-bench Verified and a 1.52-second median task time on the WindTunnel web-agent benchmark.
NVIDIA and MIT team releases Open-H3-Regenerate-2K, regenerating MiniMax H3 video drafts at up to native 4K
The Efficient-Large-Model team, made up of NVIDIA and MIT researchers, released Open-H3-Regenerate-2K weights on Hugging Face. Built on MiniMax H3, it regenerates low-resolution video drafts at 2560×1440 and up to native 3840×2160. The team reports a 3.9x speedup on a single B200 and a higher UniPercept score than the official H3-Regenerate-2K API.
Pika releases H3 Flex, an open-weight video model that generates synchronized audio in six sampling steps
Pika Labs has published Pika H3 Flex, a post-trained H3 model that generates video with synchronized audio in six sampling steps and accepts up to 12 reference inputs. Open weights run locally at native 480p and 768p, while Pika's hosted API adds 1080p, 2K and 4K output.
Edge0 releases Audio8-TTS-Realistic, a 0.6B open-weight voice cloning TTS model under Apache 2.0
Edge0 released Audio8-TTS-Realistic, a 0.6B open-weight zero-shot voice cloning text-to-speech model under Apache 2.0. It supports 23 languages and adds a final GRPO stage driven by a learned realism critic.
Comments
Loading...