Subquadratic, an AI startup based in Miami, claims that its new SubQ AI model is the first one with a 12-million-token context window and that it operates at 5% of Anthropic's Opus 4.7 costs, as per an 'X' post from co-founder Alexander Whedon on Wednesday.
In simple terms. the company has claimed that its product uses notably less compute than other AI models and is more cost-effective than other premier products, such Opus 4.7. It also said that it has a text context window of 12 million tokens (building blocks for words), this implies that it has a larger memory for remembering previous interactions than most other AIs. Anthropic's Opus 4.7 has a context window of 1 million tokens.
The firm stated that it has achieved this by reducing the amount of processing power the model spends on assessing the relationship between words, and instead focusing on what it claims to be the small amount "that actually matters".
"Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter," the post said.
It calls this design " fully sub-quadratic sparse-attention architecture".
ALSO READ: ChatGPT Makers To Launch OpenAI Smartphones By 2027? Launch Time, Top Specifications Leaked
According to reoprts, the firm has raised $29 million via seed funding from notable investors such as Tinder co-founder Justin Mateen, former SoftBank Vision Fund partner Javier Villamizar, along with early investors in Anthropic, OpenAI, Stripe, and Brex. This has reportedly raised that company's value to $500 million.
It has also unveiled three products in private beta. These include an API showing the full context window, a command-line coding agent named 'SubQ Code', and a search tool called 'SubQ Search'.
Discourse on the internet from AI experts was mixed with some believing its claims while others expressing skepticism. AI developers such as Dan Mcateer stated that early access required to investigate its claims and determine their veracity.
SubQ is either the biggest breakthrough since the Transformer...
— Dan McAteer (@daniel_mac8) May 5, 2026
> 52x faster than FlashAttention at 1mm tok context
> 20x cheaper than Opus
...or it's AI Theranos.
Requested early access so hopefully can investigate soon. https://t.co/R9tiBOZ1fu pic.twitter.com/HtYCdTYlYj
ALSO READ: Anthropic Launches Claude For Financial Services: How 10 New AI Agents Will Help Automate Work
Essential Business Intelligence, Continuous LIVE TV, Sharp Market Insights, Practical Personal Finance Advice and Latest Stories — On NDTV Profit.
