XOR
Quadrant scores
See the full quadrantScored with the public quadrant rubric (maturity × capability; bubble = adoption). Revise as evidence lands.
- Maturity5.2/10
Backed by Juspay (enterprise payments infra); clear HF release with docs; enterprise positioning implies roadmap commitment; no hosted SLA yet.
- Capability8.4/10
Choice / Score / Noul with images. Family evidence from the Gemma-based Xor 26B-A4B: JevBench v1.6.1 open board #7 (67.4), capability 72.4 (#4), Calibration 86.1, raw p50 0.27 s; Decision Index 0.3 55.95. Xor 1.2 (Qwen3.6): DI 0.3 51.15.
- Adoption28/100
Strong launch engagement (352♥ / 19.3k views); enterprise angle from Juspay/Xyne adds credibility signal; fresh release.
Vendor claims
- Open source multimodal Jev-like[Vendor claim — not independently verified]
- 260k context window (vs Jev 32k)[Vendor claim — not independently verified]
- Enterprise grade + data sovereignty[Vendor claim — not independently verified]
- Qwen3.6 based[Vendor claim — not independently verified]
XOR is an open source multimodal Jev-like model by Juspay (via their Xyne AI division) — based on Qwen3.6 with a 260k context window — not TypeSafe and positioned for enterprise-grade use cases with data sovereignty.
Text + visual classification in a Jev-style interface: pass state (text and/or images) plus typed questions, get probability distributions over options. The 260k context window is 8x Jev’s 32k limit, enabling longer document workflows.
Weights: Hugging Face juspay/xor. Announcement: @XyneAI (23 Sep 2026).
Specs
| Attribute | Value |
|---|---|
| Author | Juspay / Xyne |
| Base model | Qwen3.6 |
| Context window | 260k tokens |
| Modalities | Text + Visual |
| License | Check HF card |
| Launch | |
| Status | Open weights (Hugging Face) |
| Decision types | Choice, Score, Boolean (Jev-like interface) |
| Positioning | Enterprise-grade, data sovereignty |
Architecture
XOR follows the multimodal Jev-like pattern: text and/or visual state + typed questions → probability distribution over options. The 260k context from the Qwen3.6 backbone enables long-document and multi-image workflows that would hit Jev’s 32k ceiling.
Enterprise positioning suggests intended use cases in payments (Juspay’s domain), document processing, and workflows where data cannot leave your infrastructure.
Limits
- Benchmarks now exist, but for the newer releases — see “Juspay’s newer releases” below; the original Qwen3.6
xor1.0 was never boarded. - Day-one release — expect early rough edges; verify visual classification quality on your domains.
- No hosted SLA — self-host only; Juspay may offer enterprise deployments separately.
- Qwen3.6 backbone — inherits Qwen’s characteristics (strengths and gaps).
Juspay’s newer releases (checked 7 Oct 2026)
| Release | Base | Board results |
|---|---|---|
| Xor 26B-A4B 1.0 (1 Oct, Apache-2.0, BF16 51.6 GB; NVFP4) | Gemma-4-26B-A4B-it, two LoRA lineages souped and pulled 15% back toward the base | JevBench v1.6.1 open board #7, composite 67.4 (95% 64.6–68.7), capability 72.4 (#4), Intelligence 58.7, Calibration 86.1, raw p50 0.27 s, $0.044 per 1,000 decisions (1.36× Jev). Decision Index 0.3: 55.95 (rank 14 group) |
Xor 1.2 (juspay/xor, branch release/xor-1.2; main is still 1.1) |
Qwen3.6-35B-A3B, LoRA r16 merged | Decision Index 0.3: 51.15 (rank 23 group). Not on JevBench v1.6.1 |
Juspay’s own public-JevBench run for Xor 26B-A4B (203/231) is an UnverifiedClaim. Score change: capability 7.5 → 8.4, scored on the family’s best independently measured member (Xor 26B-A4B). Note that this is a different base from the Qwen3.6 xor this page first described. Maturity and adoption are unchanged.
Score working (7 Oct 2026): capability = types 10 × 0.30 + calibration 7 × 0.25 + latency 8 × 0.25 + accuracy 8 × 0.20 = 3.00 + 1.75 + 2.00 + 1.60 = 8.35 → 8.4.
Fit / anti-fit
Fit when you need: multimodal typed decisions with long context, self-hosted enterprise deployment, data sovereignty, document + image classification workflows.
Anti-fit when you need: hosted inference, production SLA, or workflows where Jev’s 32k context is sufficient (simpler model).
Related peers
| Model | Modalities | Context | Status |
|---|---|---|---|
| Jev-Omni | Text/Image/Audio/Video | ~30k | Open (independent) |
| XOR | Text/Visual | 260k | Open (Juspay/Xyne) |
| Jev | Text only (images via description) | 32k | Hosted (TypeSafe) |
Why it is in the catalog
First enterprise-backed open multimodal Jev-like with the largest context window in the category (260k). Fills the long-context enterprise multimodal slot distinct from independent peers like Jev-Omni.