Elorian launches Glimpse, a visual reasoning model priced at $0.60/$3 per 1M tokens
Elorian, the Palo Alto startup that raised a $55M seed at a $300M valuation, has released Glimpse as a research preview with API access at $0.60 per 1M input tokens and $3 per 1M output tokens. The model can crop, pull video frames, and run code mid-reasoning. Benchmark claims come from Elorian's own evaluations.
Elorian released Glimpse, its first AI model, on Friday as a research preview with API access priced at $0.60 per 1M input tokens and $3 per 1M output tokens, TechCrunch reported. A chat interface exists but is waitlist-only.
The Palo Alto startup raised a $55 million seed round at a $300 million valuation before shipping a product.
What Glimpse does differently
Most multimodal models encode an image once into tokens and then reason over it in text. Glimpse can return to the source material during reasoning. According to Elorian, it can:
- Crop into a specific region of an image
- Pull specific frames from a video
- Run code to measure or compare visual elements
Elorian calls this mechanism a "visual scratch space." CEO Andrew Dai, a former Google DeepMind researcher, says the motivation came from asking a frontier model who was winning a photographed board game. The model could not tell, because it could not interpret the game's iconography and symbols.
The approach is not new in itself. OpenAI added crop-and-zoom during reasoning to its o-series models in 2025. Elorian's claim is that it built the entire model around the capability rather than adding it afterward.
Pricing and positioning
| Item | Value |
|---|---|
| Input | $0.60 per 1M tokens |
| Output | $3.00 per 1M tokens |
| Context window | Not disclosed |
| Parameter count | Not disclosed |
| Training cutoff | Not disclosed |
Per TechCrunch, Glimpse is priced below mid-tier models such as Sonnet and Sol, and above lightweight models such as Haiku and Luna.
Benchmark claims
According to Elorian's own evaluations, Glimpse outperforms GPT-6 Astra on scientific chart reasoning and short-form video captioning. The company reports competitive results on other visual reasoning tasks. The reporting does not include specific scores, and the results have not been independently verified.
Elorian also claims that a small team trained Glimpse on a few hundred GPUs, and that the model is substantially smaller than the frontier systems it is benchmarked against. Exact size and compute figures were not provided.
Funding and roadmap
Elorian came out of stealth six months ago. The seed round was co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital. Nvidia and Google chief scientist Jeff Dean also participated.
Initial target customers are companies with large video archives: robotics firms, scientific labs, and media and creative companies. Longer term, Elorian wants to move from perception to action, building a model of a scene and using it to choose actions, including directly controlling robots. Dai said the company aims to build "the world's best visual reasoning model" next year, one that leads on every visual reasoning test.
What this means
Glimpse is a test of whether visual reasoning is better solved in the architecture than by scale. If Elorian's claims hold, a small model that actively re-inspects its inputs can beat much larger systems on perception-heavy tasks, at a price that makes video-heavy workloads more affordable. That matters most for robotics and scientific imaging, where fine visual detail is the bottleneck.
The evidence is thin so far. Elorian has published no independent evaluations, no context window, and no parameter count. The comparisons are against models Elorian chose. Developers evaluating Glimpse should test it on their own video and chart workloads before committing, paying particular attention to latency and token cost. Repeated crops and frame pulls can add to both, and the reporting does not say how they are billed.
Related Articles
Odyssey-3 world model enters free public preview; 14B base model, 66.1 Physics-IQ claim carries caveats
Odyssey has opened a free public research preview of Odyssey-3, an autoregressive diffusion transformer that generates interactive environments from text prompts in real time. The base model has 14 billion parameters and outputs 832 × 480 video. Odyssey's headline Physics-IQ score of 66.1 does not meet the benchmark's own requirements for a record claim.
Moondream 3 Preview lists on OpenRouter: 9B MoE vision model, 2B active, 32K context, $0.30/$2.50
Moondream 3 Preview is a vision-language model with 9B total and 2B active parameters (mixture-of-experts), a 32K-token context window and an optional reasoning mode. OpenRouter lists it at $0.30 per 1M input tokens and $2.50 per 1M output tokens, served by Prime Intellect.
Moondream 3.1: 9B MoE vision model with 2B active parameters, 32K context, $0.30/$2.50 per 1M tokens
Moondream has released Moondream 3.1, a mixture-of-experts vision-language model with 9B total and 2B active parameters and a 32K-token context window. It is listed on OpenRouter at $0.30 per 1M input tokens and $2.50 per 1M output tokens, with an optional reasoning mode.
Nvidia releases Nemotron 3.5 Super VL: open 120B multimodal MoE with 12B active parameters
Nvidia has released Nemotron 3.5 Super VL, an open multimodal Mixture-of-Experts model with 120 billion total and 12 billion active parameters, built for continuously running enterprise agents. Two compact Bolt models for local PCs and edge devices launch alongside it, though the 4 GB variant ships later this month.
Comments
Loading...