Jev vs Cloudflare Clef
Cloudflare shipped Clef in late September 2026 as a System One decision model on Workers AI: multimodal (text/JSON/images/video), Apache 2.0 open weights, Jev-API compatible. TypeSafe's Jev launched the same month. Both handle Choice/Score/Noul. Clef has open weights, and Clef-omni adds audio and video; Jev has GA, documented builds and, since 14 October 2026, image input in preview. On price, Clef-flash is now the cheapest of the lot at $0.038 per million input tokens.
Updated · by Jon Kraayenbrink
In short
- Jev: text plus PNG/JPEG images in preview, $0.042/M input (output free), GA via TypeSafe API, closed weights, madewithjev proof builds.
- Clef: multimodal (text/JSON/images/video), $0.24/M input, open weights, self-hostable. Clef-omni adds audio at $0.15/M.
- Clef-flash is the cheapest at $0.038/M since 9 October, below Jev, but its hosted context window is now 24k.
- Both: System One category, Choice/Score/Noul question types, batch up to 60–64 questions, context to 64–65k tokens.
- Category validation: the industry agrees typed decisions at speed beat generative prose for classify/route/act.
What Cloudflare announced
From the Cloudflare blog and Workers AI model docs:
- Models:
@cf/cloudflare/clefandclef-flash - Question types: noul, choice, score (System One-shaped)
- Input: multimodal — text, JSON, images, video
- Batch: up to 64 questions per call
- Context: 65,536 tokens
- Pricing: Clef $0.24 per million input tokens, unchanged on 9 October 2026 and about 1.7–2x faster. Clef-flash dropped from $0.09 to $0.038, and its hosted context window is 24k, not the 64k advertised before. Clef-omni launched at $0.15
- Clef-omni: audio, video, image and text input, open weights on Hugging Face (9 October 2026)
- Open weights: Apache 2.0 license on Hugging Face
- API compatibility: Jev-API compatible (Cloudflare blog claims vendor evaluation tables)
Sources: Workers AI Clef docs, Cloudflare blog post, Clef-omni and price change
What Jev is
TypeSafe's Jev is a decision model (System One), not a chat model that happens to pick a label. Launched September 15, 2026.
- Question types: Choice, Score, Noul (mixable in one call)
- Returns: per-option probabilities and confidence
- Input: text (string, JSON object, or array of strings), plus PNG and JPEG images in preview on 1.13 and 1.14 with the images header. No audio or video
- Box: one bounding box on an image, Jev 1.14 only (preview)
- API: documented
POST /v1/systemonewith Python and JavaScript SDKs - Pricing: $0.042 per million input tokens, output free. Images billed as input tokens, no fee per image
- Batch: ~60 questions cost about the same as one
- Limits: 64k tokens per request and 32k per question by default; 262k for both with the opt-in long-context header (preview)
Sources: TypeSafe models, image docs, Jev 1.14 post
Side by side
Facts only. Where either product has not published a detail, that is what the table says.
| TypeSafe Jev | Cloudflare Clef | |
|---|---|---|
| Announced | Sep 15, 2026 | Late Sep 2026 |
| Availability | Generally available | Generally available |
| Engine | Purpose-built System One | System One on Workers AI |
| Public contract | Documented POST /v1/systemone + SDKs | Workers AI model endpoint + HF weights |
| Question types | Choice, Score, Noul | noul, choice, score |
| Probabilities / confidence | Per-option probs + confidence | Returns probabilities (per Clef docs) |
| Input modalities | Text; PNG and JPEG images in preview | Text, JSON, images, video; Clef-omni adds audio |
| Audio and video | No | Yes (Clef-omni) |
| Bounding box answer | Box type, Jev 1.14 only (preview) | Not documented |
| Max questions per call | ~60 cost about the same as one | Up to 64 |
| Context limit | 64k per request, 32k per question; 262k opt-in (preview) | Clef 65,536 tokens; hosted Clef-flash 24k |
| Price | $0.042 / M input; output free | Clef $0.24 / M, Clef-omni $0.15 / M, Clef-flash $0.038 / M input |
| Open weights | No (closed) | Yes (Apache 2.0 on HF) |
| Self-hostable | No | Yes (HF weights) |
| Hosting | TypeSafe API only | Workers AI or self-host weights |
Price caveat: Clef prices are from Cloudflare's 9 October 2026 post and the Workers AI docs. Clef-flash is cheaper than Jev per token, but its hosted context is 24k, so a long state may not fit. Verify current pricing before hard-coding; both vendors may adjust rates.
Latency caveat: Neither vendor publishes P50/P99 latency as of early October 2026. Do not compare vendor slides to third-party measured figures without labeling methods.
What builders should do
- If you need vision (screenshots, UI state, video frames): Clef documents it, and Clef-omni adds audio. Jev reads PNG and JPEG images in preview, with no audio or video; for production, preprocess images with OCR or a vision model. See Jev and images.
- If you need self-hosted open weights: Clef publishes Apache 2.0 on Hugging Face. Jev: closed. Or see open-source alternatives (Laya, Julia 1, kev).
- If you are already on Cloudflare Workers: Clef is native; no external API wiring needed.
- If you want lowest per-token cost: Clef-flash at $0.038/M, with a 24k hosted context window. Jev is next at $0.042/M with 64k per request, then Clef-omni at $0.15/M and Clef at $0.24/M.
- If you want proof builds: Jev has madewithjev catalog with published cost/latency figures. Clef: vendor benchmarks only as of early October 2026.
Builds that show the job
Routing, classification, and agent next action — with the cost and latency each builder published.
Duncan
@ephraimduncan
Built a model router with Jev by @typesafeai. Jev decides what model fits your request best and the request is sent to that model.
XRouting and model choice
A model router on Jev
Gregor Zunic
@gregpr07
Breaking: Browser Use + Jev = Ultrafast ⚡ Findings flights took 7s and cost only $0.0039 🤯 > new action space every step > DOM state space > small LLM fallback to type (this video is at 1x speed btw) Built a tiny open source browser agent. try it below ↓
tamara
@tamarajtran
found the perfect use case for @typesafeai Jev: instant compaction in 2026, why is compaction still a summarization prompt? Jev can make it instant by scoring every tool call and dropping what’s irrelevant
XContext and memory
PickInstant compaction for Claude
Hassan
@nutlope
I used Jev to classify 1,018 AI research papers. The result: $0.08 total cost and 256ms median end-to-end latency per paper. The pipeline was: 1. Summarize each paper with DeepSeek V4 Flash 2. Send the title + summary + 24 possible topics to Jev 3. Use Jev to classify each paper 4. Visualize everything on http://1kpapers.com The summaries cost $3.99 on @togethercompute. The classifications cost $0.08 on @typesafeai. So for just over $4 of inference, I ended up with a pretty useful way to explore the top AI research papers from the past year. I think this is where things are heading: different models for different parts of the workflow, instead of using one model for everything. I’m running evals on the Jev classifications before replacing the current ones, but the site is already live: http://1kpapers.com
More on Jev use cases.
Try it today
Next steps
- Cloudflare Clef alternatives — who should use Jev instead, and when Clef still wins
- How to use Jev — your first call, SDKs, and gateways
- Jev pricing — what a decision costs across real builds
- Open-source Jev — local models that answer typed questions
- Agent prompts — wire Jev into Claude Code, Cursor, Codex
Common questions
- What is Cloudflare Clef?
- A System One decision model on Cloudflare Workers AI. Handles Choice/Score/Noul question types, multimodal (text/JSON/images/video), up to 64 questions per call, Apache 2.0 open weights on Hugging Face. Announced as Jev-API compatible.
- How is Jev different from Cloudflare Clef?
- Jev is GA via the TypeSafe API ($0.042/M input, output free), with closed weights and madewithjev proof builds. It reads text, and PNG and JPEG images in preview since Jev 1.14, but no audio or video. Clef is multimodal on Workers AI ($0.24/M input), Clef-omni adds audio and video ($0.15/M), Clef-flash is the cheapest ($0.038/M), and the weights are open, so you can self-host.
- Which is cheaper, Jev or Clef?
- It depends on the Clef model. Since Cloudflare's 9 October price change, Clef-flash is the cheapest at $0.038 per million input tokens, below Jev's $0.042. Jev is cheaper than Clef ($0.24/M) and Clef-omni ($0.15/M). Clef-flash's hosted context window is now 24k, against Jev's 64k per request. Verify current pricing before hard-coding; both APIs evolve.
- Can I self-host Jev like Clef?
- No: TypeSafe has not released Jev weights. Clef publishes Apache 2.0 weights on Hugging Face for self-hosting. For local alternatives to both, see open-source decision models (Laya, Julia 1, kev).
- Does Clef support images like OpenAI Decisions?
- Yes. Clef accepts text, JSON, images and video per Cloudflare docs, and Clef-omni adds audio. Jev reads PNG and JPEG images in preview since 14 October 2026, behind the TypeSafe-Danger-Zone-Images header, but not audio or video. OpenAI Decisions (public beta) also supports images.
- Can Jev read images like Clef?
- Yes, in preview. Jev 1.13 and 1.14 accept still PNG and JPEG images when the request sends TypeSafe-Danger-Zone-Images: true, billed as input tokens with no fee per image. TypeSafe says it is not production-ready yet. Clef-omni still leads on modalities: it takes audio and video, and Jev does not.
- Which should I use for agent routing?
- If you need audio or video: Clef-omni. If you need vision in production: Clef, or preprocess to text for Jev, since Jev's image input is a preview. If you want the lowest per-token cost and short inputs: Clef-flash at $0.038/M with a 24k hosted context. If you need more context per call at a low price: Jev at $0.042/M with 64k. If already on Workers AI: Clef is native. If you want proof builds: Jev has the madewithjev catalog.
Made with Jev is independent and not affiliated with TypeSafe AI. Every figure on this page is the one its author published, linked to where it can be checked.