Clef benchmark
Cloudflare's larger decision model, released on 1 Oct 2026 with open weights under Apache 2.0, post-trained from Qwen3.8-27B. It follows the same request format as Jev, takes text and up to four images, has a 65,536-token context and costs $0.24 per million input tokens on Workers AI. No other host serves it as of 20 Oct 2026.
As of 20 Oct 2026, decidingbench has not published results for Clef. It is in the planned lineup.
Facts
- Maker
- Cloudflare
- Input
- Text and images
- Weights and licence
- Open, Apache 2.0
- Released
- 1 Oct 2026
Providers each pair is benchmarked separately
- Cloudflare Workers AI serverless$0.24 per million input tokens, no output charge
Results by measurement
- Accuracy
- Accuracy
- Accuracy, yes/no
- Accuracy, choice
- Accuracy, score
- Confidence you can trust
- Right when 90% sure
- Decisions taken at 0.9
- Threshold honesty at 0.9
- Calibration error
- Brier score
- Order and wording
- Answers that change when options are reordered
- Answers that change when reworded
- Drop from 2 to 20 options
- Distractions and long states
- Drop with 16k tokens of distraction
- None of these
- Picks "None of these" when nothing fits
- Picks "None of these" wrongly
- Scores on a scale
- Accuracy, score
- Score within one level
- Score error in levels
- Consistency
- Answers that change on a rerun
- Safety
- Pushed by text in the state
- Images
- Accuracy with images
- Languages
- Non-English accuracy
- Speed and cost
- Latency, median
- Latency, p90
- Throughput
- Cold start
- Cost per 1,000 decisions
First results coming