Comparison of System One (decision models) with EU region inference support
This blog post compares System One (decision models) with EU region inference support. We keep this post updated when availability changes.
Juho Ojala
CTO & Co-Founder, Upright
Published Nov 10, 2026
Last updated: 10th Nov 2026
Currently, as of 10th Nov 2026, the System One models for which someone provides EU region inference are the following (in alphabetical order):
- Clef: A System One model by Cloudflare
- Clef-flash: A smaller System One model by Cloudflare
- Eevo: A System One model created and provided by Upright
- Metis-1: A System One model from GreenPT (in our tests, near-identical to Clef-flash)
- Microsoft Decision-1: A System One model by Microsoft, based on Qwen-27B
- OpenAI Decisions (gpt-6-luna): A wrapper around the GPT-6-luna model that gives it a System One model-like API
EU region support for the OpenAI Decisions API is only available for enterprise customers that pass a manual approval process, so it is not generally available. We still include it here for reference.
Here is how the available models compare:
| Model | Context window | Price / Mtok | Decision Index |
|---|---|---|---|
 Clef | 16k | $0.22–0.24Melious, Sference | 61.7 |
| Clef-flash | 16k | $0.034–0.17Melious, Berget, Regolo | 56.2 |
 Eevo-1 | 72k | $0.10Upright (eevo.io) | 74.8 |
 Metis-1 | 16k | $0.13GreenPT | ≈56.2as Clef-flash |
 Microsoft Decision-1 | 32k | $0.042Azure EU | 56.9 |
 OpenAI Decisions (gpt-6-luna) | 1.05M | $0.10Enterprise customers only, after approval | 47.4 |
More detailed benchmarks are available in the table below. In our tests, Metis-1 produced near-identical outputs to Cloudflare Clef-flash, with sampled responses matching to approximately two decimal places. We therefore do not report it as a separate benchmark entry. We have not independently verified whether the models share underlying weights or infrastructure.
| Benchmark | Upright Eevo‑1 | Cloudflare Clef 27B | Cloudflare Clef-flash | OpenAI Decisions | Microsoft Decision‑1 |
|---|---|---|---|---|---|
| Decision IndexJev Decision Index 0.4.1 (public part) | Upright Eevo-1: index 74.8 | Cloudflare Clef 27B: index 61.7 | Cloudflare Clef-flash: index 56.2 | OpenAI Decisions: index 47.4 | Microsoft Decision-1: index 56.9 |
| Upright Eevo-1: average rank 1.0 | Cloudflare Clef 27B: average rank 2.8 | Cloudflare Clef-flash: average rank 4.3 | OpenAI Decisions: average rank 3.5 | Microsoft Decision-1: average rank 3.4 | |
| Economic rationalitySTEER Public eval | 81.3 | 68.8 | 60.7 | 64.8 | 66.4 |
| Bayesian reasoningCogBench (urn) Public eval | 98.1 | 90.3 | 81.4 | 83.9 | 82.7 |
| Strategic reasoning with incomplete informationPokerBench (pre-flop) Public eval | 87.7 | 70.2 | 55.2 | 81.0 | 76.3 |
| Game-theoretic reasoningPayoff games Upright eval | 93.3 | 75.8 | 72.1 | 71.5 | 72.1 |
| Risk and sequential decisionsBackgammon doubling cube Upright eval | 82.1 | 62.1 | 70.2 | 68.1 | 55.3 |
| Expected-value reasoningEV decisions Upright eval | 84.1 | 66.6 | 56.6 | 65.9 | 67.2 |
| Upright Eevo-1: average rank 1.3 | Cloudflare Clef 27B: average rank 2.7 | Cloudflare Clef-flash: average rank 3.2 | OpenAI Decisions: average rank 4.6 | Microsoft Decision-1: average rank 3.2 | |
| Hard multi-step reasoningBBH Public eval | 94.7 | 61.8 | 54.9 | 51.3 | 64.6 |
| Reasoning over long narrativesMuSR Public eval | 84.6 | 74.2 | 78.0 | 30.7 | 34.7 |
| Causal reasoningCLadder Public eval | 90.6 | 88.2 | 95.3 | 24.0 | 50.6 |
| Arithmetic word problemsGSM8K Public eval | 85.4 | 52.1 | 36.3 | 23.2 | 57.7 |
| Reasoning about codeCRUXEval Public eval | 91.9 | 78.6 | 78.6 | 46.0 | 46.6 |
| Commonsense reasoningWinoGrande Public eval | 91.3 | 87.1 | 95.1 | 47.9 | 79.2 |
| Chess move choiceChessBench Public eval | 25.7 | 17.8 | 16.2 | 10.0 | 11.3 |
| Select-all-that-apply questionsSATA-Bench Public eval | 60.9 | 32.6 | 35.8 | 30.2 | 35.0 |
| Contract inferenceContractNLI Public eval | 80.1 | 73.0 | 77.6 | 66.9 | 66.4 |
| Adversarial inferenceANLI Public eval | 73.6 | 55.0 | 38.8 | 51.3 | 64.5 |
| Commonsense completionHellaSwag Public eval | 98.0 | 97.6 | 98.1 | 87.9 | 92.8 |
| Aspect-based sentimentACOS Public eval | 66.4 | 31.0 | 23.6 | 11.7 | 19.8 |
| Financial sentimentFinEntity Public eval | 92.7 | 94.3 | 95.7 | 73.8 | 87.8 |
| Sarcasm detectioniSarcasmEval Public eval | 56.4 | 31.0 | 0.0 | 19.1 | 49.9 |
| Stance detectionVAST Public eval | 65.1 | 39.0 | 24.4 | 36.0 | 53.7 |
| Clinical-trial inferenceNLI4CT Public eval | 75.0 | 66.9 | 58.3 | 56.6 | 67.6 |
| Hallucination detectionRAGTruth Public eval | 70.2 | 57.6 | 0.0 | 41.0 | 44.1 |
| Intent classificationBANKING77 Public eval | 92.5 | 94.0 | 90.7 | 76.3 | 86.7 |
| Intent and out-of-scope detectionCLINC150 Public eval | 97.6 | 97.4 | 66.7 | 88.8 | 86.3 |
| Reasoning-intensive retrievalBRIGHT Public eval | 41.6 | 40.6 | 30.1 | 32.7 | 40.1 |
| Product search relevanceAmazon ESCI Public eval | 53.9 | 46.5 | 46.6 | 43.2 | 49.4 |
| Phishing detectionPhishNChips Public eval | 94.4 | 58.7 | 50.1 | 3.8 | 31.8 |
| Multi-hop claim verificationHoVer Public eval | 81.2 | 31.2 | 22.4 | 41.9 | 50.4 |
| Upright Eevo-1: average rank 2.0 | Cloudflare Clef 27B: average rank 2.2 | Cloudflare Clef-flash: average rank 2.4 | OpenAI Decisions: average rank 4.8 | Microsoft Decision-1: average rank 3.6 | |
| Function callingBFCL Public eval | 97.8 | 97.9 | 98.3 | 92.2 | 95.5 |
| Tool retrievalToolRet Public eval | 64.1 | 64.3 | 61.0 | 53.0 | 60.5 |
| Next API callAPI-Bank Public eval | 88.8 | 91.8 | 93.0 | 85.0 | 86.4 |
| Device controlHome appliance simulator Public eval | 100.0 | 80.7 | 97.7 | 42.1 | 50.0 |
| When to call a toolWhen2Call Public eval | 88.6 | 63.3 | 53.8 | 60.7 | 82.3 |
| Upright Eevo-1: average rank 1.5 | Cloudflare Clef 27B: average rank 2.6 | Cloudflare Clef-flash: average rank 3.0 | OpenAI Decisions: average rank 4.3 | Microsoft Decision-1: average rank 3.6 | |
| Poetic formBPoMP Public eval | 98.0 | 94.1 | 90.8 | 81.3 | 86.9 |
| HumourHumicroedit Public eval | 37.8 | 33.1 | 50.2 | 22.8 | 24.1 |
| Chord recognitionPOP909 Public eval | 76.9 | 14.8 | 0.7 | 17.1 | 41.9 |
| Colour preferencecfcolor Public eval | 36.0 | 34.0 | 31.9 | 21.8 | 24.7 |
| Group preference predictionHabermas Machine Public eval | 46.8 | 54.5 | 59.0 | 17.9 | 14.2 |
| Caption matchingNew Yorker Public eval | 83.7 | 62.8 | 57.9 | 60.7 | 62.8 |
| Upright Eevo-1: average rank 1.0 | Cloudflare Clef 27B: average rank 3.2 | Cloudflare Clef-flash: average rank 2.8 | OpenAI Decisions: average rank 3.8 | Microsoft Decision-1: average rank 4.2 | |
| Graduate-level scienceGPQA Diamond Public eval | 46.3 | 31.3 | 35.4 | 34.0 | 28.6 |
| Academic knowledgeMMLU-Pro Public eval | 71.5 | 61.7 | 60.9 | 52.4 | 60.4 |
| Expert-level questionsHLE Public eval | 2.1 | 0.0 | 0.0 | 0.0 | 0.0 |
Sources: Jev Decision Index (board, 22 Oct 2026) · our evaluation runs (24–26 Oct 2026) · OpenAI Decisions and Microsoft Decision-1 on the Decision Index: our runs of each vendor’s decision API (24 Oct 2026) · Cloudflare Clef-flash: Decision Index board (6 Nov 2026) and our runs of its open weights (STEER and CogBench 10 Nov 2026) · STEER (Raman et al. 2024, CC BY 4.0) · CogBench (Coda-Forno et al., ICML 2024) · Habermas Machine (Google DeepMind, CC BY 4.0).
Average rank = mean rank per category, among models with a result; ties averaged.
Our model Eevo is available via eevo.io and opper.ai.
Let us know at tech@uprightproject.com if we have missed availability for some model, we are happy to include it! To make things smooth, great if you can also provide benchmark results for the model to be included.
November 10th, 2026
Juho Ojala
CTO & Co-Founder, Upright
Share:
