Jev vs Xor 26B-A4B
Apache-2.0 Gemma-4-26B-A4B in bf16, with a separate Xor 1.2 line.
Jev 1.13.0
TypeSafe · hosted System One model
77.1Capability · unranked reference
Intelligence 63.6 · calibration 90.6.
Xor 26B-A4B
Juspay · Apache-2.0 Gemma-4-26B-A4B in bf16, with a separate Xor 1.2 line
72.4Open-weights Capability #4
Capability #4 at 72.4. Intelligence 58.7, calibration 86.1.
Key differences
- Where it runs JevHosted APIXor 26B-A4BSelf-hosted bf16, author’s SGLang serving bundle
- Licence JevProprietary, hostedXor 26B-A4BApache-2.0
- Capability (v1.6.1) Jev77.1 · referenceXor 26B-A4B72.4 · rank #4
- Intelligence / calibration Jev63.6 / 90.6Xor 26B-A4B58.7 / 86.1
- This row’s base JevTypeSafe Jev 1.13.0Xor 26B-A4BGemma-4-26B-A4B, bf16
- A different Juspay model Jev—Xor 26B-A4BXor 1.2 is Qwen3.6-35B-A3B and is not the ranked row
JevBench v1.6.1, same method
Capability is the mean of intelligence and calibration, ranked only inside Benchmark Heaven’s cost and median-latency caps. Jev 1.13.0 stays on the board as the reference row and is not given a rank. A model can score above that reference and still be left off the numbered list when its estimated cost is too high. Full table ↗
Capability axes, higher is better
Cost $0.044 per 1,000 decisions (estimated). Median latency 0.3 s. Jev’s reference row is $0.032 per 1,000 decisions (not published as a list price), median 0.2 s. Both dollar figures are the benchmark’s, except Jev’s, which Benchmark Heaven marks as not a published list price.
Which Xor?
The JevBench v1.6.1 row is Xor 26B-A4B: Juspay’s Gemma-4-26B-A4B checkpoint in bf16, Apache-2.0, served with their SGLang bundle and a compatibility server. Capability is 72.4, rank #4 among eligible open-weights systems. Jev’s reference is 77.1, with higher intelligence (63.6 vs 58.7) and higher calibration (90.6 vs 86.1).
Juspay also publishes Xor 1.2, a post-train of Qwen3.6-35B-A3B (about 35B total, about 3B active) that serves /v1/systemone. That is a different base and a different release. Numbers from the Xor 1.2 card do not describe the 26B-A4B row on this page.
Read this before you switch
- Mixing Xor 1.2’s Qwen 35B-A3B card with the Gemma 26B-A4B board row will send you to the wrong model.
- Jev leads both Capability axes that this rank averages.
- Estimated cost is $0.044 per 1,000 decisions against Jev’s $0.032, and median latency is 0.3 s against 0.2 s.
When to use which
Choose Jev if
- You want the higher Capability reference (77.1 vs 72.4) without standing up SGLang.
- Calibration and intelligence both matter, and Jev leads each.
- You do not want to track which Xor checkpoint a tutorial is about.
Choose Xor 26B-A4B if
- You want Juspay’s Apache-2.0 26B-A4B weights on your own GPU.
- You already run their serving bundle and want the compatibility server in front of it.
- You have read the 26B-A4B decision-index notes and you mean that checkpoint, not Xor 1.2.
FAQ
Is Xor better than Jev?
Not on this board. Xor 26B-A4B is Capability #4 at 72.4. Jev’s reference is 77.1, with intelligence 63.6 vs 58.7 and calibration 90.6 vs 86.1.
Is Xor 1.2 the same model?
No. Xor 1.2 is a post-train of Qwen3.6-35B-A3B. The ranked JevBench row is Gemma-4-26B-A4B in bf16. Use the checkpoint hash in the decision-index dataset before you compare scores.
Does Xor speak the Jev API?
Xor 1.2’s card documents a TypeSafe-compatible /v1/systemone server. The 26B-A4B release uses the same compatibility-server pattern. Either way it is Juspay’s model, not Jev.