SubQ Snap lands on Baseten

// 1h agoINFRASTRUCTURE

SubQ Snap lands on Baseten

Subquadratic’s long-context decision model is now available through Baseten’s inference platform. It supports up to 384K tokens of multimodal context and returns structured probabilities for agent routing, tool selection, checking, and stopping. [SubQ](https://subq.ai/introducing-subq-snap) [Baseten](https://www.baseten.co/library/publisher/subquadratic/)

// ANALYSIS

This is a meaningful distribution win for specialized agent infrastructure: SubQ Snap targets the bounded decisions that otherwise consume expensive frontier-model calls.

–Typed outputs make routing and tool arbitration easier to threshold and integrate programmatically
–A 384K-token context window can reduce lossy chunking across repositories, documents, and agent histories
–Baseten availability lowers deployment friction for teams already using its inference stack
–Subquadratic reports stronger decision quality than Jev on several enterprise and long-context evaluations
–Developers should validate latency, quotas, pricing, and preview stability on Baseten before production rollout

// TAGS

subq-snapllminferencelong-contextmultimodalagentapihosted-service

DISCOVERED

1h ago

2026-11-02

PUBLISHED

2h ago

2026-11-02

RELEVANCE

8/ 10

AUTHOR

subquadratic

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

UPDATE16m ago

Exa Adds 3.2M Episodes to Search API

Exa now indexes 3.2 million podcast episodes across 119,000 shows, with full speaker-labeled transcripts and timestamps. The dataset is searchable through Exa’s existing API, enabling agents to discover, summarize, and recommend episodes based on what was actually said.

UPDATE22m ago

ADE Puts iPhone Duo Agents In Pocket

ADE’s iPhone app gives developers mobile access to agent chats, terminals, simulators, and approvals, including workflows for testing Apple’s foldable iPhone Duo. Its synced, multi-agent setup makes remote supervision practical for developers working across devices.

BENCHMARK27m ago

GPT-6 Astra Performs LLM Weight Surgery

A new demonstration claims GPT-6 Astra can directly interpret and manipulate another LLM’s weights, producing a sixfold improvement on math tasks and roughly threefold gains on other capabilities. The result suggests frontier models may be moving from using models to actively modifying them.