Benchmarks
More
OverviewBenchmaxxingCompareChartsCost vs CapabilityProviders per ModelProvider explorerGatewaysEU & SovereignJev-class modelsImage JevBenchAbout
JevBench by Benchmark Heaven · v1.4.2.2 · individual system

swanOne (blockbrain, Qwen3.8-Flash-Next NVFP4)

system-one-open · by blockbrain

JevBench v1.4.2.2 score

33.6

Rank #36 of 91 ranked systems.

29.7 points behind Jev 1.13.0's 63.3.

Where it sits among the 91 ranked systems. The marked tick is Jev 1.13.0 (63.3).

Published axes

intelligence
52.6
calibration
71.0
speed
82.5
cost
38.6

Against Jev 1.13.0

Four radars compare this fixed pair across the score axes, accuracy per tier, and accuracy by family on the hard tier and sealed set. Further out is better on every spoke.

  • A: swanOne — system-one-open · Score 33.6 (#36)
  • B: Jev 1.13.0 — Jev (TypeSafe, closed) · Score 63.3 (#4)

The four score axes

Radar: the four score axes, two systemsThe four score axes, swanOne vs Jev 1.13.0. Intelligence: 52.6 vs 53.1; Calibration: 71.0 vs 76.3; Speed: 82.5 vs 83.3; Cost: 38.6 vs 52.0.50100Intelligence52.6 · 53.1Calibration71.0 · 76.3Speed82.5 · 83.3Cost38.6 · 52.0
0–100, the values in the table. A label-only system has no calibration (counted as 0).

Accuracy per tier, incl. sealed

Radar: accuracy per tier, incl. sealed, two systemsAccuracy per tier, incl. sealed, swanOne vs Jev 1.13.0. Easy: 100% vs 100%; Standard: 96% vs 99%; Judge: 93% vs 95%; Hard: 75% vs 74%; Sealed: 39% vs 37%.50100Easy100% · 100%Standard96% · 99%Judge93% · 95%Hard75% · 74%Sealed39% · 37%
Share correct per tier; Sealed = the 308 private decisions, aggregate only.

Current question set by family (hard + sealed)

Radar: current question set by family (hard + sealed), two systemsCurrent question set by family (hard + sealed), swanOne vs Jev 1.13.0. Ambiguous / abstain: 63% vs 43%; Judge: 51% vs 54%; Long policy: 53% vs 44%; Multi-hop: 68% vs 64%; Probability: 44% vs 63%; Temporal / numeric: 28% vs 28%; Trade-off: 58% vs 55%; Routing: 100% vs 100%; Trap / adversarial: 88% vs 83%; Paraphrase: 29% vs 64%; Safety judge: 50% vs 38%.50100Ambiguous /abstain63% · 43%Judge51% · 54%Long policy53% · 44%Multi-hop68% · 64%Probability44% · 63%Temporal /numeric28% · 28%Trade-off58% · 55%Routing100% · 100%Trap /adversarial88% · 83%Paraphrase29% · 64%Safety judge50% · 38%
Share correct per family across the 220 hard-tier decisions (public and held out) and the 308 sealed decisions of v1.4, pooled; Routing is hard-tier only, Paraphrase and Safety judge sealed only.

Sealed set by family

Radar: sealed set by family, two systemsSealed set by family, swanOne vs Jev 1.13.0. Ambiguous / abstain: 54% vs 30%; Judge: 34% vs 34%; Long policy: 33% vs 28%; Multi-hop: 53% vs 45%; Paraphrase: 29% vs 64%; Probability: 25% vs 50%; Safety judge: 50% vs 38%; Temporal / numeric: 25% vs 29%; Trade-off: 42% vs 38%; Trap / adversarial: 67% vs 42%.50100Ambiguous /abstain54% · 30%Judge34% · 34%Long policy33% · 28%Multi-hop53% · 45%Paraphrase29% · 64%Probability25% · 50%Safety judge50% · 38%Temporal /numeric25% · 29%Trade-off42% · 38%Trap /adversarial67% · 42%
Share correct within each sealed family — system-level aggregates; the items stay private.

Availability and evidence

Openness
Code and weights marked open in the published row
License note
patches Apache-2.0, shim MIT; weights under the Qwen licence (LICENSE-NOTICE.md)
Cost evidence
estimated; the board’s row disclosure contains the published basis.
Endpoint condition
our evaluator-owned Lium GPU pod (RTX PRO 6000), offline read-only container, author's server on loopback
Note on this row
Draft vocabulary derived via AGPL-3.0 generator (provenance recorded separately). Submitter consulted all 231 public tasks and swept temperature over 111 public hard tasks. Score corrected during review from pooled-534 ECE/latency to hard-tier/242-item block. blockbrain-ai/swanone-recipe abcca2e789316472e58d4e824b61c04f419c3ba5: Mia-AiLab/Qwen3.8-Flash-Next-NVFP4 rev 925d7be6c14c6c9442ef83e8f05b5a3c39304f69 on vllm/vllm-openai:qwen38-flash-next@sha256:0aea3024… with MiaAI Lab's nine patched vLLM files (hash-verified) and the author's shim (option letters in the benchmark's order, the model's own renormalised letter mass, no temperature), H100.md serve command, here on an RTX PRO 6000 Blackwell. The shim sends its own system prompt; structured state is rendered with json indent=1. Offline self-hosted inference of all 842 decisions (534 frozen v1.2 + 308 sealed v1.4) in a network-disabled, read-only container on an evaluator-owned Lium RTX PRO 6000 Blackwell 96 GB pod, through JevBench's unchanged typesafe adapter against the author's own server on loopback; no operator endpoint; no golds were exposed. Latency is the serial request wall time on the standard+judge items with the self-hosted adjustment. Cost is a labelled estimate: OpenRouter qwen/qwen3.8-flash list price $0.15/M input (same underlying Flash-Next weights; the NInfer Flash-Next precedent), read 2026-09-24, over the server's own usage.input_tokens; it is not a GPU bill.
Published source
https://github.com/blockbrain-ai/swanone-recipe

From the public v1.4.2.2 aggregate. Scores and ranks can change when a new release is published.

Read the full board and published method. The overall score is a composite, not raw accuracy.