# Cloudflare Decision Index 0.2.1: API-Bank

> Published Cloudflare and community decision-model configurations.

Canonical page: https://benchlm.ai/benchmarks/clefdecisionindex021task3

- Category: [Decision Models](/decision-models)
- Last updated: October 1, 2026 capture

Evaluation results published by [Cloudflare and upstream community contributors](https://clef-evals.workers-ai-mle.workers.dev/). undefined

The source combines Cloudflare self-reported Clef runs with upstream community configurations. Each row retains its source engine and evidence label; model names alone do not establish equivalent serving settings.

The 0.2.1 composite uses 38 panel benchmarks grouped into five weighted areas. It applies chance correction; ForecastBench uses its Brier transformation. Additional task views retain the source metric and panel-inclusion metadata. The snapshot lists 48 task metrics, while the launch refers to 43 evaluations including latency rows.

Clef and Clef-Flash each cover 36 of 38 index tasks; HLE and iSarcasmEval are missing. Missing values are not inserted as zero. The composite is retained as published without recomputing it or filling gaps.

Community results use the upstream September 28 snapshot, which declares one NVIDIA RTX PRO 6000. Cloudflare run hardware is not established by this artifact. Latency comparisons across these configurations do not isolate model speed. The October 1 date marks capture, not evaluation.

These views remain separate from JevBench and canonical provider benchmark results. They do not enter overall or category rankings. Equal scores do not establish statistical ties.

- [Cloudflare evaluation site](https://clef-evals.workers-ai-mle.workers.dev/)
- [Published result artifact](https://clef-evals.workers-ai-mle.workers.dev/data/leaderboard.json)
- [Upstream Decision Index](https://huggingface.co/spaces/multimodalart/jev-decision-index)
- [Clef launch](https://blog.cloudflare.com/clef-decision-models/)
- [Clef model card](https://huggingface.co/Cloudflare/clef)
- [Clef-Flash model card](https://huggingface.co/Cloudflare/clef-flash)

## About Cloudflare Decision Index 0.2.1: API-Bank

- Year: 2026

Cloudflare Decision Index 0.2.1: API-Bank is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (64 configurations)

| Rank | Model | Configuration | Creator | Score |
|------|-------|---------------|---------|-------|
| 1 | [Clef-flash](https://huggingface.co/Cloudflare/clef-flash) | Cloudflare self-reported; Routing head + LoRA; 36/38 index tasks; median 38.8 ms; p95 122.4 ms | Cloudflare | 93.1% |
| 2 | [Clef](https://huggingface.co/Cloudflare/clef) | Cloudflare self-reported; Routing head + LoRA; 36/38 index tasks; median 209.3 ms; p95 238.6 ms | Cloudflare | 91.9% |
| 3 | [Jev](https://clef-evals.workers-ai-mle.workers.dev/) | Upstream community result; Hosted API (closed); 38/38 index tasks; median 524.1 ms; p95 536 ms | TypeSafe | 88.2% |
| 4 | [Eikos-27B [FP8]](https://huggingface.co/caiovicentino1/Eikos-27B-FP8) | Upstream community result; LoRA; 38/38 index tasks; median 129.7 ms; p95 509.8 ms | — | 86.0% |
| 5 | [simple-jev · Qwen3.8-27B (featherless)](https://github.com/featherless-ai/simple-jev) | Upstream community result; inference technique; 38/38 index tasks; median 373 ms; p95 2796.8 ms | — | 85.4% |
| 6 | [Decider chat · Gemma-4-31B](https://github.com/Mapika/decider) | Upstream community result; inference technique; 38/38 index tasks; median 108.5 ms; p95 1114.7 ms | — | 85.0% |
| 7 | [reflex 27B [FP8 · wide choice]](https://github.com/kshetrajna12/reflex) | Upstream community result; inference technique; 38/38 index tasks; median 108.3 ms; p95 1239.7 ms | — | 85.0% |
| 8 | [AutoJev-27B](https://huggingface.co/denis-pplx/autojev-27b) | Upstream community result; full fine-tune; 38/38 index tasks; median 101.4 ms; p95 1052.5 ms | — | 84.1% |
| 9 | [Jebadiah 27B](https://huggingface.co/frontier-infra/jebadiah-27b) | Upstream community result; LoRA; 38/38 index tasks; median 110.4 ms; p95 591.1 ms | — | 84.1% |
| 10 | [Jevfire](https://github.com/kikoncuo/jevfire) | Upstream community result; inference technique; 38/38 index tasks; median 89.6 ms; p95 447.5 ms | — | 84.1% |
| 11 | [JoshuaSP diffusiongemma (open-jev)](https://github.com/JoshuaSP/open-jev) | Upstream community result; inference technique; 38/38 index tasks; median 260.4 ms; p95 537.3 ms | — | 83.7% |
| 12 | [Surogate Rune 26B-A4B v3 [bf16]](https://huggingface.co/surogate/rune-26b-a4b-GGUF) | Upstream community result; full fine-tune; 38/38 index tasks; median 120.5 ms; p95 682.8 ms | — | 83.3% |
| 13 | [Bespoke Nimble 9B v2](https://huggingface.co/bespokelabs/Bespoke-Nimble-9B-v2) | Upstream community result; LoRA; 38/38 index tasks; median 77.3 ms; p95 621.4 ms | — | 80.5% |
| 14 | [Decider chat · Qwen3.6-27B](https://github.com/Mapika/decider) | Upstream community result; inference technique; 38/38 index tasks; median 83.6 ms; p95 965.6 ms | — | 79.7% |
| 15 | [lev](https://huggingface.co/interfaze-ai/lev) | Upstream community result; LoRA + head; 38/38 index tasks; median 70.8 ms; p95 687.7 ms | — | 79.1% |
| 16 | [JPT-9B](https://huggingface.co/kirp/jpt-9b) | Upstream community result; LoRA; 38/38 index tasks; median 148.6 ms; p95 461.9 ms | — | 78.7% |
| 17 | [Decision 1.0 Nox](https://huggingface.co/llm-semantic-router/Decision-1.0-Nox) | Upstream community result; head / adapter; 38/38 index tasks; median 53.2 ms; p95 242.1 ms | — | 77.6% |
| 18 | [Intern-Decision-4B](https://huggingface.co/internlm/Intern-Decision-4B) | Upstream community result; full fine-tune; 38/38 index tasks; median 44.2 ms; p95 115.4 ms | — | 76.0% |
| 19 | [Hopper (G) 1.2](https://huggingface.co/HopitAI/hopper-g) | Upstream community result; LoRA; 38/38 index tasks; median 23.3 ms; p95 238.2 ms | — | 75.6% |
| 20 | [Decision 1.0 Lux](https://huggingface.co/llm-semantic-router/Decision-1.0-Lux-9B) | Upstream community result; head / adapter; 38/38 index tasks; median 51 ms; p95 347.5 ms | — | 74.0% |
| 21 | [Winnow-E4B [Q8_0]](https://huggingface.co/EldanRing/Winnow-E4B) | Upstream community result; LoRA; 38/38 index tasks; median 45 ms; p95 188.3 ms | — | 73.0% |
| 22 | [djev](https://github.com/Davipar/djev-dev) | Upstream community result; inference technique; 38/38 index tasks; median 84.4 ms; p95 211.2 ms | — | 70.7% |
| 23 | [JPT-4B](https://huggingface.co/kirp/jpt-4b) | Upstream community result; LoRA; 38/38 index tasks; median 139.1 ms; p95 405.7 ms | — | 68.7% |
| 24 | [JevK5](https://huggingface.co/alibiserikbay/JevK5) | Upstream community result; LoRA; 38/38 index tasks; median 22 ms; p95 226.3 ms | — | 68.5% |
| 25 | [Winnow-12B [Q8_0]](https://huggingface.co/EldanRing/Winnow-12B) | Upstream community result; LoRA; 38/38 index tasks; median 72.5 ms; p95 356.2 ms | — | 67.3% |
| 26 | [NeoHorse-Jev-4B](https://huggingface.co/TokenRhythm/NeoHorse-Jev-4B) | Upstream community result; head / adapter; 38/38 index tasks; median 49 ms; p95 273.9 ms | — | 66.1% |
| 27 | [Decision 1.0 Sol](https://huggingface.co/llm-semantic-router/Decision-1.0-Sol) | Upstream community result; head / adapter; 38/38 index tasks; median 38.1 ms; p95 101 ms | — | 62.0% |
| 28 | [Bosun v3.1 1.7B](https://huggingface.co/Hanno-Labs/bosun-v3.1-1.7b) | Upstream community result; LoRA + head; 38/38 index tasks; median 40.3 ms; p95 173.5 ms | — | 60.8% |
| 29 | [Decider 4B](https://huggingface.co/Mapika/decider-4b) | Upstream community result; full fine-tune; 38/38 index tasks; median 12.6 ms; p95 188.5 ms | — | 60.4% |
| 30 | [razorback16 openjev diffusiongemma (NVFP4, vLLM) [one read, NVFP4]](https://github.com/vllm-project/vllm/pull/57250) | Upstream community result; inference technique; 38/38 index tasks; median 37.7 ms; p95 117.8 ms | — | 57.3% |
| 31 | [Decider 35B-A3B [NVFP4]](https://huggingface.co/Mapika/decider-35b-a3b-nvfp4) | Upstream community result; full fine-tune; 38/38 index tasks; median 101.4 ms; p95 210.9 ms | — | 56.5% |
| 32 | [Kev 9B](https://huggingface.co/jaredpalmer/kev-9b) | Upstream community result; LoRA + head; 38/38 index tasks; median 51.4 ms; p95 187.9 ms | — | 56.3% |
| 33 | [Jet v6.2](https://huggingface.co/michaljach/jet) | Upstream community result; LoRA; 38/38 index tasks; median 46.8 ms; p95 371.1 ms | — | 54.3% |
| 34 | [Bosun v3.1 0.6B](https://huggingface.co/Hanno-Labs/bosun-v3.1-0.6b) | Upstream community result; LoRA + head; 38/38 index tasks; median 38.7 ms; p95 107.3 ms | — | 53.1% |
| 35 | [Jev-Omni](https://huggingface.co/akhilaaa3/Jev-Omni) | Upstream community result; LoRA + head; 38/38 index tasks; median 54.9 ms; p95 627.9 ms | — | 52.4% |
| 36 | [this-that 1.2](https://huggingface.co/flock-io/this-that-model-1.2) | Upstream community result; full fine-tune; 38/38 index tasks; median 44.3 ms; p95 57.2 ms | — | 52.4% |
| 37 | [Kev 4B](https://huggingface.co/jaredpalmer/kev-4b) | Upstream community result; LoRA + head; 38/38 index tasks; median 52.1 ms; p95 141.1 ms | — | 51.2% |
| 38 | [open-jev (pngwn)](https://huggingface.co/pngwn/system-one-qwen3.5-4b-scorer-v2b) | Upstream community result; LoRA; 38/38 index tasks; median 112.3 ms; p95 724.1 ms | — | 48.4% |
| 39 | [Jobe](https://github.com/MantisShrimpdev/jobe) | Upstream community result; inference technique; 38/38 index tasks; median 53.5 ms; p95 2910.3 ms | — | 46.1% |
| 40 | [openvons](https://github.com/genai-craft/openvons) | Upstream community result; inference technique; 38/38 index tasks; median 21.2 ms; p95 113.3 ms | — | 43.5% |
| 41 | [Kev 0.8B](https://huggingface.co/jaredpalmer/kev-0.8b) | Upstream community result; LoRA + head; 38/38 index tasks; median 41.6 ms; p95 95.9 ms | — | 42.9% |
| 42 | [Decision 1.0 Eos](https://huggingface.co/llm-semantic-router/Decision-1.0-Eos-0.8B) | Upstream community result; head / adapter; 38/38 index tasks; median 39.7 ms; p95 84.3 ms | — | 36.6% |
| 43 | [Decider 2B [FP8]](https://huggingface.co/Mapika/decider-2b) | Upstream community result; full fine-tune; 38/38 index tasks; median 8.1 ms; p95 134.9 ms | — | 31.7% |
| 44 | [JPT-0.8B](https://huggingface.co/kirp/jpt-0.8b) | Upstream community result; LoRA; 38/38 index tasks; median 117.5 ms; p95 341.2 ms | — | 30.1% |
| 45 | [MoJev](https://huggingface.co/MoLeMo-Lab/mojev) | Upstream community result; head / adapter; 38/38 index tasks; median 46.4 ms; p95 55.1 ms | — | 29.1% |
| 46 | [Intern-Decision-2B](https://huggingface.co/internlm/Intern-Decision-2B) | Upstream community result; full fine-tune; 38/38 index tasks; median 33.5 ms; p95 58 ms | — | 28.5% |
| 47 | [Lavoir](https://huggingface.co/moganai/lavoir) | Upstream community result; full fine-tune; 38/38 index tasks; median 20 ms; p95 49.5 ms | — | 28.3% |
| 48 | [Intern-Decision-0.8B](https://huggingface.co/internlm/Intern-Decision-0.8B) | Upstream community result; full fine-tune; 38/38 index tasks; median 33.7 ms; p95 39.2 ms | — | 21.1% |
| 49 | [Julia 1](https://huggingface.co/SupersonicLabs/Julia-1) | Upstream community result; full fine-tune; 38/38 index tasks; median 5.8 ms; p95 15.3 ms | — | 18.3% |
| 50 | [Decision 1.0 Kai](https://huggingface.co/llm-semantic-router/Decision-1.0-Kai) | Upstream community result; head / adapter; 38/38 index tasks; median 30.4 ms; p95 106.8 ms | — | 15.9% |
| 51 | [GLiNER 2.5 base](https://huggingface.co/fastino/gliner2.5-base-v1) | Upstream community result; full fine-tune; 38/38 index tasks; median 14.1 ms; p95 62.1 ms | — | 15.6% |
| 52 | [Decision 1.0 Lex](https://huggingface.co/llm-semantic-router/decision-1.0-lex) | Upstream community result; head / adapter; 38/38 index tasks; median 29.8 ms; p95 106 ms | — | 14.0% |
| 53 | [Laya](https://huggingface.co/convaiinnovations/laya) | Upstream community result; full fine-tune; 38/38 index tasks; median 5.8 ms; p95 222.5 ms | — | 11.5% |
| 54 | [Lumma-Fev-0.1B](https://huggingface.co/FrontiersMind/Lumma-fev-0.1b) | Upstream community result; full fine-tune; 38/38 index tasks; median 20.9 ms; p95 112.5 ms | — | 8.3% |
| 55 | [GLiNER2.5-Decide](https://huggingface.co/fastino/GLiNER2.5-Decide) | Upstream community result; full fine-tune; 38/38 index tasks; median 23.3 ms; p95 228.6 ms | — | 7.7% |
| 56 | [CLM-v0.1-8B](https://huggingface.co/Contrastive-LM/CLM-v0.1-8B) | Upstream community result; full fine-tune; 38/38 index tasks; median 46.8 ms; p95 130.9 ms | — | 6.1% |
| 57 | [Qwen-2.5-1B-RLCD](https://huggingface.co/harshatheg/Qwen-2.5-1B-RLCD) | Upstream community result; inference technique; 38/38 index tasks; median 43.8 ms; p95 75.2 ms | — | 4.3% |
| 58 | [Lumma-Fev-0.6B](https://huggingface.co/FrontiersMind/Lumma-fev-0.6b) | Upstream community result; LoRA; 38/38 index tasks; median 21.3 ms; p95 219.7 ms | — | 4.1% |
| 59 | [jeff](https://github.com/logan-markewich/jeff) | Upstream community result; inference technique; 38/38 index tasks; median 21.8 ms; p95 67.6 ms | — | 3.1% |
| 60 | [LFM2.5-350M-RLCD](https://huggingface.co/notnotsamuel/LFM2.5-350M-RLCD) | Upstream community result; full fine-tune; 38/38 index tasks; median 26.7 ms; p95 174.3 ms | — | 2.4% |
| 61 | [GLiNER 2.5 small](https://huggingface.co/fastino/gliner2.5-small-v1) | Upstream community result; full fine-tune; 38/38 index tasks; median 13.6 ms; p95 35.2 ms | — | 2.0% |
| 62 | [LFM2.5-2.6B-RLCD](https://huggingface.co/monotykamary/LFM2.5-2.6B-RLCD) | Upstream community result; full fine-tune; 38/38 index tasks; median 39.3 ms; p95 95.4 ms | — | 1.8% |
| 63 | [GLiNER 2.5 multilingual](https://huggingface.co/fastino/gliner2.5-multi-v1) | Upstream community result; full fine-tune; 38/38 index tasks; median 14 ms; p95 92.4 ms | — | 1.6% |
| 64 | [system-one-gemma](https://github.com/akash-kamat/system-one-gemma) | Upstream community result; LoRA + head; 38/38 index tasks; median 31.9 ms; p95 748 ms | — | 1.2% |

## FAQ

### What does Cloudflare Decision Index 0.2.1: API-Bank measure?

Published Cloudflare and community decision-model configurations.

### Which model leads the published Cloudflare Decision Index 0.2.1: API-Bank snapshot?

Clef-flash currently leads the published Cloudflare Decision Index 0.2.1: API-Bank snapshot with a score of 93.1%.

### How many models are evaluated on Cloudflare Decision Index 0.2.1: API-Bank?

The October 1, 2026 capture contains 64 AI models.

### Does Cloudflare Decision Index 0.2.1: API-Bank affect BenchLM's overall score?

Not directly. Cloudflare Decision Index 0.2.1: API-Bank is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.