Is Sage2 state of the art?Pending benchmarks.Announced Nov 2, 2026; no independent results yet.
Levanto Labs' runtime model for developers, with two modes: Extract turns natural language into typed structured fields, and Decide answers yes/no, choice, scale, sort and tag questions, including over a stream that it reads as the context grows. Hosted API only.
The details so far
- Lab
- Levanto Labs
- Announced
- Nov 2, 2026
- API
- Live · levanto-sage-2
- Open weights
- Unknown · None announced
- Modes
- Extract (natural language in, typed structured fields out) and Decide (yes/no, choice, scale, sort and tags)
- Model ID
- Both modes report levanto-sage-2 in API responses
- Streaming
- Decide can read a conversation or document as it grows over a WebSocket stream (/v1/stream): 1 to 32 questions and up to 16,384 tokens of text and questions per session
- Inputs
- Text; images in beta
- Options
- Reasoning, web-search grounding, batch calls and a SystemOne-compatible decision endpoint
- Access
- Levanto API (sage.levanto.ai) on monthly plans billed per input token, output token and search, with a free plan that includes 20M tokens. Also on NEAR AI and, for agents, through MPP with Stripe cards and Tempo stablecoins
- Parameters
- Not disclosed
- Benchmark
- Extract Bench v0.1 (1,844 documents from 12 public datasets), built and maintained by Levanto and open-sourced under Apache 2.0
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: Extract is 166x cheaper than Opus 5.5 and beats Cloudflare's Clairic.
Lab claim: on Levanto's Extract Bench v0.1, a Sage 2 preview (rc9, fast) scores 77.8 field accuracy and 72.2 on fields not in the document against 75.3 and 59.0 for Clairic.
Lab claim: Decide is 3x faster on progressive context because it reads the context as it grows, with streamed decisions as low as 50 ms; extraction takes about 150 ms of server time on short documents.
Sources
LABLevanto Labs announcement on X: Introducing Sage2x.com/levantolabs/status/2117304026911908020 · Nov 2, 2026LABLevanto Labs on X: Extract Bench resultsx.com/levantolabs/status/2117304032255455505 · Nov 2, 2026LABSage2 model page (levanto.ai)levanto.ai/modelLABLevanto docs: streamingdocs.levanto.ai/streamREPOlevantolabs/extract-bench on GitHubgithub.com/levantolabs/extract-benchTracking continues until an independent board lists it.