Specifications side by side
Read from each model's page, so a change there updates this table. A blank field is dated to the day that model's page was last checked.
| Specification | Jev | Clef |
|---|---|---|
| Vendor | TypeSafe AI | Cloudflare |
| Status | Generally available | Generally available |
| Access | Hosted API | Hosted API |
| Question types | Choice, Score, Noul | Choice, Score, Noul |
| Input price | $0.042 per 1M tokens | $0.24 per 1M tokens |
| Output price | Free | Free |
| Total context | 64,000 tokens | 65,536 tokens |
| State budget | 32,000 tokens | Not published as of 14 Nov 2026 |
| Latency | 70 to 500 ms1 | 38.8 to 209.3 ms2 |
| Max Choice options | 255 | Not published as of 14 Nov 2026 |
| Score levels | Up to 10 | Not published as of 14 Nov 2026 |
| Rate limit | Set per organization and shown in the TypeSafe developer console. TypeSafe publishes no fixed figures, checked on 2026-11-14. Going over the token or request limit returns HTTP 429, and higher limits are available on custom and enterprise plans. | Not published on the Workers AI model pages as of 2026-10-02. |
| SDKs | Python, TypeScript | None listed |
| Licence | Not published as of 14 Nov 2026 | Apache-2.0 |
| Jev Decision Index, Full score | 60.24 scorev0.4.1, 12 Nov 2026 | 54.92 scorev0.4.1, 12 Nov 2026 |
| Jev Decision Index, Public score | 60.7 scorev0.4.1, 12 Nov 2026 | 61.71 scorev0.4.1, 12 Nov 2026 |
| Jev Decision Index, Expected calibration error | 0.0799 ECEv0.4.1, 12 Nov 2026 | 0.0395 ECEv0.4.1, 12 Nov 2026 |
| JevBench, Capability Score | 75 scorev1.8.10, 16 Oct 2026 | 75.3 scorev1.8.10, 12 Oct 2026 |
| Microsoft 36-benchmark comparison, Average accuracy | 82.3 %vendor-run9 Oct 2026 | Not published as of 14 Nov 2026 |
| Microsoft 36-benchmark comparison, Calibration score | 93.7 scorevendor-run9 Oct 2026 | Not published as of 14 Nov 2026 |
| Artificial Analysis Decision Index, Index score | 66.2 scorev2.1, 12 Nov 2026 | 60.3 scorev2.1, 12 Nov 2026 |
| Artificial Analysis Decision Index, Expected calibration error | 0.0536 ECEv2.1, 12 Nov 2026 | 0.0709 ECEv2.1, 12 Nov 2026 |
| Artificial Analysis Decision Index, Cost per 1M tasks | 156 USDv2.1, 12 Nov 2026 | 863 USDv2.1, 12 Nov 2026 |
| Artificial Analysis Decision Index, Time per task | 92 msv2.1, 12 Nov 2026 | 354 msv2.1, 12 Nov 2026 |
| CLINC150+OOS, macro-F1 | Not published as of 14 Nov 2026 | 97.43 scorevendor-run1 Oct 2026 |
| BANKING77, macro-F1 | Not published as of 14 Nov 2026 | 94.2 scorevendor-run1 Oct 2026 |
| Docs and console | docs.typesafe.ai, Console | developers.cloudflare.com |
- 1 Jev: 70ms to 500ms end-to-end is TypeSafe's own figure from the launch post, as is the 3 to 329 seconds it compares frontier LLMs against and the resulting 40x to 200x multiple.
- 2 Clef: Vendor-run, from Cloudflare's launch post, across the 43 evals it ran: median latency 209.3ms for Clef and 38.8ms for Clef-flash, against 524.1ms for Jev, 51.4ms for Kev-9B and 5.8ms for Laya.
Examples in the directory
638 Jev examples
The top 3 by stars and views. All Jev examples.

Jev Ultrafast (Browser Use)
Browser Use's fast browser agent where Jev picks the operation and the DOM element in a single request; a small LLM is only invoked to write text when typing is required. Demoed on a Google Flights search.
19k starsFlight search 7.1 s

openjev
Attempt to run a Jev-style System One decision model locally on a single RTX 3090. Companion site at openjev.com. The repository was renamed to TheoLeeCJ/SemIf on GitHub, and the old openjev URL now redirects to it.
4.1k starsJSON vs direct 5.21x slowerPeak throughput 20.03 dec/s
jev-trader
Makes one AI trade decision every Monad block on the Kuru MON-USDC market using Jev. The public deployment runs in dry-run mode with a mock model.
2.2k starsLoop, mock model p50 100 msDecision window ~300 ms/block
7 Clef examples
The top 3 by stars and views. All Clef examples.

Introducing Clef: our open-source decision models, and new RL fine-tuning platform
Cloudflare's launch post for Clef and Clef-flash, two decision models on Workers AI that accept the Jev request format and open weights under Apache 2.0. It covers the architecture, Cloudflare's own benchmark tables against Jev, Kev and Laya, and a new reinforcement learning service.

Decision Model Leaderboard
Cloudflare's live leaderboard for decision models on the Jev Decision Index suite, with Clef, Clef-flash, Jev, Kev-9B and Laya plotted by score and latency. Cloudflare built and ran it, so treat the ranking as vendor-run.
Cloudflare/clef: open weights on Hugging Face
Apache 2.0 weights for Cloudflare's 27B Clef decision model, post-trained from Qwen3.8-27B, that reads text and images. The smaller 9B Clef-flash is published next to it as Cloudflare/clef-flash.
Shared use cases
Patterns both Jev and Clef list on their pages.
- Support inbox triage
- Intent and model routing
- Confidence-gated actions
- Agent routing and skill selection
The short answer
Clef is the closest thing to a drop-in neighbour that Jev has. Cloudflare trained it, hosts it on Workers AI, and says it is “fully Jev-API compatible”, so a request with a state and typed Noul, Choice and Score questions runs on either. Clef also reads images, the 27B model has twice Jev’s context window as Cloudflare states it, and both sizes ship open weights under Apache 2.0.
On price the two sizes fall either side of Jev. The 27B Clef costs nearly six times as much, and the 9B Clef-flash a little less, on Workers AI and on OpenRouter alike. Jev has the longer public record. Most comparisons between the two come from Cloudflare’s launch post, which Cloudflare ran itself, so read those as one side’s results. Jev wins several of them. Two outside boards have run both. The Jev Decision Index 0.3 (multimodalart’s community board on Hugging Face) has Clef ahead on public benchmarks and Jev ahead once private tests are counted. The Artificial Analysis Decision Index, launched on 13 October 2026, ranks Jev 1.14 11th of 22 decision models at 66.2 on its version 2.1, and Clef 19th at 60.3, as of 14 November 2026. Version 2.1 has no Clef-flash row.
Jev vs Clef at a glance
| Jev | Clef | Clef-flash | |
|---|---|---|---|
| Maker | TypeSafe AI | Cloudflare | Cloudflare |
| Size and base | Not published | 27B, frozen Qwen3.8-27B backbone | 9B, frozen Qwen3.5-9B backbone |
| Weights | Closed, API only | Apache 2.0 on Hugging Face | Apache 2.0 on Hugging Face |
| Hosted API | TypeSafe, OpenRouter, Cloudflare | Workers AI as @cf/cloudflare/clef; OpenRouter as cloudflare/clef |
Workers AI as @cf/cloudflare/clef-flash; OpenRouter as cloudflare/clef-flash |
| Question types | Choice, Score, Noul | Choice, Score, Noul | Choice, Score, Noul |
| Input | Text | Text, JSON, images, video | Text, JSON, images, video |
| Context window | 64,000 tokens per request, 32,000 for state plus the longest question; both 262k with an opt-in header; Cloudflare’s post gives 32,000 | 65,536 tokens on Workers AI and on OpenRouter | 24,576 tokens on Workers AI and on OpenRouter |
| Questions per request | Not checked on this page | 1 to 64 | 1 to 64 |
| Input price per million tokens | $0.042, output free | $0.24 on Workers AI and OpenRouter, $0.192 through DeepInfra on OpenRouter, output free | $0.038 on Workers AI and OpenRouter, $0.019 through DeepInfra on OpenRouter, output free |
| Median latency (Cloudflare’s run, 1 October) | 524.1ms | 209.3ms | 38.8ms |
| p95 latency (Cloudflare’s run, 1 October) | 536.0ms | 238.6ms | 122.4ms |
| Median latency after the 9 October speedup (Cloudflare) | Not re-measured | 152ms at about 800 tokens, 305ms at about 3,400, 1,635ms at about 16,000 | Not re-measured |
Sources: TypeSafe’s models page and Cloudflare’s launch post, checked 2 October 2026. Clef’s prices and contexts come from the Workers AI model pages and OpenRouter’s endpoints API, checked 12 November 2026. There Cloudflare and PrimeIntellect charge the Workers AI prices, and DeepInfra serves both models in fp8 at a discount. On 10 October PrimeIntellect’s context was 16,384 tokens for both models and Cloudflare’s OpenRouter endpoint gave Clef-flash 65,536. Earlier checks found Clef-flash at $0.09 on Workers AI on 2 October, and PrimeIntellect charging $0.042 for Clef and $0.021 for Clef-flash on 9 October. The first two latency rows are the medians and 95th percentiles from Cloudflare’s 43-eval run at launch. The third row is from its 9 October post, which reports Clef only, by input size. TypeSafe’s own claim for Jev is 70ms to 500ms end to end.
What each model does
Both models read one state and answer a set of typed questions with probabilities. Neither writes text. Jev’s three question types are covered in Choice, Score and Noul, and Clef uses the same three with the same field names.
Clef adds an optional images array of up to four PNG, JPEG or WebP images, which the Workers AI docs call a Clef extension to the System One API. It also requires a model field set to clef or clef-flash. A client written for Jev needs those two changes plus a Cloudflare account and token.
How Clef was built
Cloudflare’s post says each model keeps a frozen Qwen backbone and trains a routing head and rank-256 low-rank adapters on top. At inference it makes one prefill pass and then scores every valid answer in parallel, so no text is generated token by token. Training used label-smoothed cross-entropy plus a Brier loss for calibration, synthetic data, and a second stage Cloudflare calls Reinforcement Learning for Calibrated Decisions. Cloudflare does not say that stage matches TypeSafe’s RLCD, which is still unpublished.
Jev’s architecture is not published, so there is no like-for-like description to set beside this. How Jev is built covers what is known.
What Cloudflare’s benchmarks show
All of these are vendor-run: Cloudflare chose the tasks, ran the models and published the tables. Nobody else has reproduced these tables. The independent run in the next section uses its own suite.
On 10 tasks it picked from the Jev Decision Index:
| Task and metric | Clef | Clef-flash | Jev |
|---|---|---|---|
| BFCL, case exact | 98.47 | 98.76 | 95.75 |
| ToolRet, nDCG@10 | 69.19 | 66.43 | 65.28 |
| API-Bank, accuracy | 91.93 | 93.11 | 88.19 |
| Home appliances, case exact | 82.95 | 97.73 | 52.27 |
| When2Call, accuracy | 72.37 | 65.58 | 80.97 |
| BANKING77, macro-F1 | 94.20 | 90.93 | 79.74 |
| CLINC150+OOS, macro-F1 | 97.43 | 66.77 | 89.27 |
| BRIGHT, nDCG@10 | 45.91 | 39.26 | 47.52 |
| Amazon ESCI, macro-F1 | 57.48 | 57.39 | 55.21 |
| PhishNChips, accuracy | 79.60 | 75.05 | 62.55 |
Clef scores higher than Jev on eight of the ten. Jev is ahead on When2Call and on BRIGHT. Clef-flash is below Jev on three tasks, and by the widest margin on CLINC150+OOS.
On TypeSafe’s own workflow evals:
| Workflow | Clef | Clef-flash | Jev |
|---|---|---|---|
| Invoice processing | 64.7 | 57.1 | 61.8 |
| Customer service | 76.3 | 77 | 76.0 |
| Security incidents | 62.9 | 61.7 | 61.7 |
| Agent trace observability | 68.5 | 69.8 | 71.6 |
Clef is ahead of Jev on three of the four, by 0.3 of a point on customer service, and behind on agent trace observability. Cloudflare also says Clef is currently the leader when evaluated against the Jev Decision Index, and publishes a live leaderboard for it.
What the Jev Decision Index shows
The public Jev Decision Index added Clef and Clef-flash in edition 0.3 on 7 October 2026; the figures here are from edition 0.4.1, in the file generated on 12 November 2026. It is a third-party board run by one person, with the same suite and rules for every entry. Since edition 0.4 its headline Full score takes 60% from private tests of the same skills and 40% from private tasks in new domains, and the public benchmarks no longer count toward it.
| Jev Decision Index 0.4.1 | Clef | Clef-flash | Jev 1.14 |
|---|---|---|---|
| Full score | 54.92 | 44.43 | 60.24 |
| Public benchmarks only | 61.71 | 56.15 | 60.70 |
| Private tests of the same skills | 53.11 | 44.01 | 59.28 |
| Private tasks from new domains | 45.06 | 35.00 | 48.26 |
| Calibration error, lower is better | 0.040 | 0.025 | 0.080 |
Clef scores above Jev 1.14 on the public benchmarks. It falls to 54.92 on the Full score, 104th of 301 rows, because it scores lower on the private tests. In the 0.3.1 file of 10 October 2026, Clef scored 53.08 and Clef-flash 47.61 on the Full score. Both Clef models have lower calibration error than Jev. On neither score is Clef the leader that Cloudflare’s own leaderboard shows.
Speed, cost and hosting
Cloudflare reports a median of 209.3ms for Clef and 38.8ms for Clef-flash against 524.1ms for Jev, across the 43 evals it ran on 1 October. The post does not say how Jev was called or from where. Cloudflare’s 9 October post says it then made the hosted Clef faster: the median at about 800 tokens fell from 262ms to 152ms, at about 3,400 tokens from 616ms to 305ms and at about 16,000 tokens from 2,721ms to 1,635ms. It gave no new figure for Clef-flash or Jev, and the two sets of Clef figures use different inputs. The Jev median is above TypeSafe’s own upper figure of 500ms, so the gap may partly reflect the network path. Measure both from your own region.
Price depends on the size. On Workers AI, Clef costs $0.24 per million input tokens and Clef-flash $0.038, against $0.042 for Jev, checked on 10 October 2026. Clef-flash cost $0.09 at launch. None of the three charges for output. OpenRouter serves both Clef models through Cloudflare and PrimeIntellect at the Workers AI prices, and through DeepInfra in fp8 for $0.192 and $0.019, checked on 12 November 2026. Because the weights are Apache 2.0, you can also run either Clef model on your own GPU, which Jev does not allow.
Cloudflare says it does not read, store or train on requests or responses unless you opt into its fine-tuning service. Jev pricing covers Jev’s published figures.
When to use which
- You need to classify images. Clef takes up to four per request. Jev reads images only behind an opt-in header that TypeSafe says is not ready for production.
- You need to classify audio or video. Cloudflare’s third model, Clef-omni, launched on 9 October 2026 at $0.15 per million input tokens, takes both as well as images. Jev takes neither.
- You send long states. Clef’s window is 65,536 tokens on Workers AI, against 32,000 for Jev’s state plus the longest question by default, or 262k with an opt-in header. Clef-flash on Workers AI takes 24,576 tokens in all.
- Cost per call is the main limit. Clef-flash, at $0.038 per million input tokens, is a little cheaper than Jev at $0.042. Clef, at $0.24, costs nearly six times as much. OpenRouter charges the same through Cloudflare and PrimeIntellect, and less through DeepInfra.
- Latency is the main limit. Cloudflare’s numbers favour Clef-flash by a wide margin, but they are its own. Test in your region.
- Your tasks look like When2Call or agent traces. Cloudflare’s tables put Jev ahead there. Test on your own data.
- You want weights you control. Only Clef offers them. See Jev alternatives for other open options.
FAQ
Is Clef a Jev alternative?
Yes, for the request format. Cloudflare says Clef is fully Jev-API compatible, so the same state and Noul, Choice and Score questions run on both. Moving means changing the endpoint, adding a model field and rerunning your own evals. It is not identical. Clef takes images, and its context window and price depend on the size and the provider.
Is Clef better than Jev?
Cloudflare’s own benchmarks say Clef is higher on most of the tasks it chose, and lower on When2Call, BRIGHT and agent trace observability. On the independent Jev Decision Index 0.4.1, Clef beats Jev 1.14 on the public benchmarks, 61.71 to 60.7, and trails it on the Full score, built from private tests, 54.92 to 60.24. Which is better depends on your data, so run both on a sample of it.
Is Clef free or open source?
The weights are free under Apache 2.0 on Hugging Face, for both Clef and Clef-flash. The hosted versions are not free. Clef costs $0.24 per million input tokens and Clef-flash $0.038, on Workers AI and on OpenRouter, checked on 10 October 2026. Running the weights yourself costs the GPU time, and the 27B model needs a large GPU.
Does Clef read images?
Yes. The Workers AI docs say Clef reads the state as text, JSON, images or video, and accept up to four PNG, JPEG or WebP images per request through an images field. Cloudflare calls that field a Clef extension to the System One API. Cloudflare’s launch post says Jev reads text only. Since 14 October 2026 TypeSafe accepts images on Jev behind an opt-in header.
Is Clef faster than Jev?
In Cloudflare’s own launch run, yes: a median of 209.3ms for Clef and 38.8ms for Clef-flash against 524.1ms for Jev. Cloudflare later sped Clef up, to a 152ms median at about 800 tokens. It does not say how it called Jev, and TypeSafe claims 70ms to 500ms end to end. Treat both sets of numbers as vendor claims and time them yourself.
Examples

Introducing Clef: our open-source decision models, and new RL fine-tuning platform
Cloudflare's launch post for Clef and Clef-flash, two decision models on Workers AI that accept the Jev request format and open weights under Apache 2.0. It covers the architecture, Cloudflare's own benchmark tables against Jev, Kev and Laya, and a new reinforcement learning service.

Decision Model Leaderboard
Cloudflare's live leaderboard for decision models on the Jev Decision Index suite, with Clef, Clef-flash, Jev, Kev-9B and Laya plotted by score and latency. Cloudflare built and ran it, so treat the ranking as vendor-run.
Cloudflare/clef: open weights on Hugging Face
Apache 2.0 weights for Cloudflare's 27B Clef decision model, post-trained from Qwen3.8-27B, that reads text and images. The smaller 9B Clef-flash is published next to it as Cloudflare/clef-flash.

Introducing Clef-omni with full multimodality, plus a faster Clef and a cheaper Clef-flash
Cloudflare's launch post for Clef-omni, a 30B-A3B mixture-of-experts decision model that reads audio, video, images and text, at $0.15 per million input tokens. It also cuts Clef-flash to $0.038 with a 24K hosted context, and gives Cloudflare's own benchmark and latency tables.

Jev on Cloudflare Workers AI
Cloudflare Workers AI model catalog page for TypeSafe Jev, documenting the state plus questions request shape and the Noul, Choice and Score answer types. Examples cover ticket routing, refund review and risk assessment.