Glimpse
An AI model that keeps looking back at the image while it reasons.
About
Glimpse is the first model from Elorian, the Palo Alto startup founded by former Google DeepMind researcher Andrew Dai, released October 16, 2026 as a research preview with API access (chat interface waitlist-only). Its core idea is a 'visual scratch space': instead of encoding an image once and reasoning in text, Glimpse returns to the source mid-thought — cropping regions, pulling video frames, running code to measure or compare what it sees. Elorian says a small team trained it on a few hundred GPUs and reports it beats OpenAI's GPT-6 Astra on scientific chart reasoning and short-form video captioning, while openly showing its losses elsewhere. Pricing is $0.60 per million input tokens and $3 per million output tokens.
What it can do
- Visual scratch space: re-inspects images and video mid-reasoning (provider-claimed)
- Crop into regions, densify video frames, track objects across frames (provider-claimed)
- Beats GPT-6 Astra on scientific chart reasoning and short-form video captioning per company evals (provider-claimed)
- Research-preview API at $0.60/1M input, $3/1M output tokens (provider-claimed)
Setup
- Request API access at elorian.ai (research preview; chat is waitlist-only).
- Obtain API credentials once approved.
- Send images or video with prompts; Glimpse re-inspects the visuals mid-reasoning and returns measurements and comparisons.
Prerequisites
API access approval (research preview); chat interface is waitlist-only.
Use cases
Works with
Trust score — 5/10
Scored 7/10 (METRICS tier, metrics-v1). Evidence: TechCrunch exclusive Oct 16, 2026; founder is ex-Google DeepMind; $55M seed at $300M valuation (April 2026); company publishes a results table that includes its own losses to GPT-6 Astra — unusually candid; official URL verified live (200, Oct 2026). Strengths: genuinely novel mechanism, honest disclosure of model size and gaps. Docked for: (1) all benchmark figures are company-run, no independent verification; (2) API is research preview, chat waitlist-only — no hands-on validation possible yet; (3) closed weights and unannounced training data. License: proprietary API (research preview).
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 3.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · First model from Elorian (Andrew Dai, ex-Google DeepMind); released Oct 16, 2026 as research preview with API access, chat waitlist-only
- · $55M seed at $300M valuation (April 2026); company claims small team trained it on a few hundred GPUs
- · Company-reported wins over GPT-6 Astra on chart reasoning/video captioning, with own losses disclosed
- · Pricing $0.60/1M input, $3/1M output tokens
- · manual review: all benchmark figures are company-run with no independent replication; API is research preview so no hands-on validation possible yet; closed weights and unannounced training data
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.