# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/clef-our-open-source-decision-models-49923692

# Clef: Open\-weight decision models, and new RL fine\-tuning platform

604 points · 213 comments

[Full discussion](<https://news.ycombinator.com/item?id=49923692>)

[Read original](<https://blog.cloudflare.com/clef-decision-models/>)

Category: [Models & Products](<https://hacksnap.live/?category=models-products>)

## Skept-o-meter & Hotness

Skept\-o\-meter: High\. Estimated from 10 comments\.

1 comment for the summary\.

Peak rank: \#1

Time in Top 10: 24\.0 hours

Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\.

151 recorded rank observations from 2026\-10\-01T18:01:44\.046011\+00:00 to 2026\-10\-10T23:01:00\.95027\+00:00\.

Hotness — latest 151 recorded Hacksnap ranks:

2026\-10\-01T18:01:44\.046011\+00:00: rank \#10

2026\-10\-01T19:01:47\.420738\+00:00: rank \#5

2026\-10\-01T20:01:39\.585083\+00:00: rank \#3

2026\-10\-01T21:01:45\.872085\+00:00: rank \#1

2026\-10\-01T22:01:37\.214899\+00:00: rank \#2

2026\-10\-01T23:04:11\.395288\+00:00: rank \#2

2026\-10\-02T08:02:26\.819307\+00:00: rank \#2

2026\-10\-02T09:02:51\.51649\+00:00: rank \#2

2026\-10\-02T10:02:12\.346099\+00:00: rank \#2

2026\-10\-02T11:02:22\.794376\+00:00: rank \#2

2026\-10\-02T12:01:14\.681578\+00:00: rank \#2

2026\-10\-02T13:02:29\.511591\+00:00: rank \#2

2026\-10\-02T14:01:40\.936738\+00:00: rank \#2

2026\-10\-02T15:02:30\.225476\+00:00: rank \#2

2026\-10\-02T16:02:47\.060695\+00:00: rank \#2

2026\-10\-02T17:02:31\.351493\+00:00: rank \#2

2026\-10\-02T18:01:22\.271899\+00:00: rank \#39

2026\-10\-02T19:01:23\.681417\+00:00: rank \#38

2026\-10\-02T20:01:33\.801055\+00:00: rank \#40

2026\-10\-02T21:02:55\.453316\+00:00: rank \#43

2026\-10\-02T22:01:50\.233139\+00:00: rank \#44

2026\-10\-02T23:01:20\.268516\+00:00: rank \#42

2026\-10\-03T08:01:30\.186188\+00:00: rank \#43

2026\-10\-03T09:01:57\.91467\+00:00: rank \#40

2026\-10\-03T10:00:55\.732001\+00:00: rank \#39

2026\-10\-03T11:01:22\.538596\+00:00: rank \#39

2026\-10\-03T12:01:08\.789384\+00:00: rank \#40

2026\-10\-03T13:00:35\.273559\+00:00: rank \#40

2026\-10\-03T14:00:48\.425606\+00:00: rank \#40

2026\-10\-03T15:01:04\.472963\+00:00: rank \#40

2026\-10\-03T16:01:56\.56175\+00:00: rank \#41

2026\-10\-03T17:01:36\.651036\+00:00: rank \#38

2026\-10\-03T18:01:58\.24207\+00:00: rank \#38

2026\-10\-03T19:01:19\.196589\+00:00: rank \#37

2026\-10\-03T20:00:57\.313759\+00:00: rank \#35

2026\-10\-03T21:01:04\.44439\+00:00: rank \#33

2026\-10\-03T21:20:01\.585706\+00:00: rank \#32

2026\-10\-03T22:00:15\.503126\+00:00: rank \#32

2026\-10\-03T23:00:55\.584832\+00:00: rank \#32

2026\-10\-04T08:01:00\.327459\+00:00: rank \#31

2026\-10\-04T09:00:59\.321023\+00:00: rank \#32

2026\-10\-04T10:00:42\.975336\+00:00: rank \#32

2026\-10\-04T11:01:03\.76577\+00:00: rank \#32

2026\-10\-04T12:01:45\.307985\+00:00: rank \#32

2026\-10\-04T13:00:49\.176267\+00:00: rank \#32

2026\-10\-04T14:00:56\.952543\+00:00: rank \#32

2026\-10\-04T15:00:54\.739637\+00:00: rank \#32

2026\-10\-04T16:00:51\.403054\+00:00: rank \#32

2026\-10\-04T17:00:43\.166713\+00:00: rank \#32

2026\-10\-04T18:00:46\.315535\+00:00: rank \#31

2026\-10\-04T19:00:54\.768841\+00:00: rank \#31

2026\-10\-04T20:00:51\.070442\+00:00: rank \#32

2026\-10\-04T21:00:28\.550532\+00:00: rank \#31

2026\-10\-04T22:01:36\.217651\+00:00: rank \#33

2026\-10\-04T23:01:09\.400767\+00:00: rank \#33

2026\-10\-05T08:01:51\.387003\+00:00: rank \#32

2026\-10\-05T09:01:40\.528475\+00:00: rank \#31

2026\-10\-05T10:00:56\.290772\+00:00: rank \#31

2026\-10\-05T11:01:05\.919518\+00:00: rank \#31

2026\-10\-05T12:00:56\.590881\+00:00: rank \#30

2026\-10\-05T13:01:05\.197156\+00:00: rank \#32

2026\-10\-05T14:01:04\.085718\+00:00: rank \#32

2026\-10\-05T15:00:50\.814571\+00:00: rank \#32

2026\-10\-05T16:02:09\.54139\+00:00: rank \#31

2026\-10\-05T17:00:35\.336076\+00:00: rank \#31

2026\-10\-05T18:00:20\.445491\+00:00: rank \#31

2026\-10\-05T19:02:43\.604975\+00:00: rank \#32

2026\-10\-05T20:01:09\.777459\+00:00: rank \#32

2026\-10\-05T21:01:47\.650671\+00:00: rank \#33

2026\-10\-05T22:02:59\.93099\+00:00: rank \#33

2026\-10\-06T08:02:11\.210914\+00:00: rank \#33

2026\-10\-06T08:02:14\.354697\+00:00: rank \#33

2026\-10\-06T09:01:21\.688765\+00:00: rank \#35

2026\-10\-06T10:00:58\.047773\+00:00: rank \#35

2026\-10\-06T11:00:41\.689929\+00:00: rank \#36

2026\-10\-06T12:00:41\.7705\+00:00: rank \#37

2026\-10\-06T13:01:24\.509519\+00:00: rank \#37

2026\-10\-06T14:00:26\.950817\+00:00: rank \#37

2026\-10\-06T15:00:49\.066414\+00:00: rank \#38

2026\-10\-06T16:00:58\.166747\+00:00: rank \#40

2026\-10\-06T17:01:17\.009511\+00:00: rank \#40

2026\-10\-06T18:01:43\.791429\+00:00: rank \#41

2026\-10\-06T19:00:46\.048609\+00:00: rank \#40

2026\-10\-06T20:01:14\.687294\+00:00: rank \#40

2026\-10\-06T21:02:13\.414382\+00:00: rank \#39

2026\-10\-06T22:00:40\.926271\+00:00: rank \#38

2026\-10\-06T23:03:26\.867445\+00:00: rank \#39

2026\-10\-07T08:01:41\.339804\+00:00: rank \#40

2026\-10\-07T09:02:38\.026986\+00:00: rank \#41

2026\-10\-07T10:01:35\.069531\+00:00: rank \#41

2026\-10\-07T11:02:07\.770084\+00:00: rank \#41

2026\-10\-07T12:01:14\.996697\+00:00: rank \#40

2026\-10\-07T13:00:47\.45912\+00:00: rank \#39

2026\-10\-07T14:01:18\.307739\+00:00: rank \#39

2026\-10\-07T15:01:57\.541362\+00:00: rank \#41

2026\-10\-07T16:01:28\.474022\+00:00: rank \#40

2026\-10\-07T17:01:15\.036908\+00:00: rank \#41

2026\-10\-07T18:00:53\.300079\+00:00: rank \#40

2026\-10\-07T19:01:49\.456657\+00:00: rank \#41

2026\-10\-07T20:01:59\.032487\+00:00: rank \#41

2026\-10\-07T21:05:36\.655392\+00:00: rank \#43

2026\-10\-07T22:02:27\.105241\+00:00: rank \#44

2026\-10\-07T23:01:17\.452894\+00:00: rank \#43

2026\-10\-08T08:03:55\.083808\+00:00: rank \#43

2026\-10\-08T09:03:58\.68898\+00:00: rank \#41

2026\-10\-08T10:04:05\.364235\+00:00: rank \#42

2026\-10\-08T11:02:04\.410307\+00:00: rank \#41

2026\-10\-08T12:02:36\.39618\+00:00: rank \#41

2026\-10\-08T13:03:09\.206401\+00:00: rank \#41

2026\-10\-08T14:03:30\.113384\+00:00: rank \#42

2026\-10\-08T15:02:54\.444758\+00:00: rank \#40

2026\-10\-08T16:03:13\.067807\+00:00: rank \#39

2026\-10\-08T17:03:47\.92465\+00:00: rank \#38

2026\-10\-08T18:01:50\.360431\+00:00: rank \#38

2026\-10\-08T19:02:55\.719274\+00:00: rank \#39

2026\-10\-08T20:01:36\.359022\+00:00: rank \#39

2026\-10\-08T21:03:34\.000719\+00:00: rank \#39

2026\-10\-08T22:01:13\.459732\+00:00: rank \#38

2026\-10\-08T23:01:25\.318392\+00:00: rank \#37

2026\-10\-09T08:01:53\.314021\+00:00: rank \#38

2026\-10\-09T09:02:59\.93426\+00:00: rank \#38

2026\-10\-09T10:02:22\.652527\+00:00: rank \#37

2026\-10\-09T11:02:04\.807282\+00:00: rank \#37

2026\-10\-09T12:02:16\.769295\+00:00: rank \#37

2026\-10\-09T13:02:30\.35034\+00:00: rank \#39

2026\-10\-09T14:01:34\.215134\+00:00: rank \#38

2026\-10\-09T15:01:27\.564392\+00:00: rank \#37

2026\-10\-09T16:02:06\.561056\+00:00: rank \#38

2026\-10\-09T17:03:06\.673648\+00:00: rank \#40

2026\-10\-09T18:01:02\.851052\+00:00: rank \#40

2026\-10\-09T19:01:11\.250291\+00:00: rank \#40

2026\-10\-09T20:01:32\.91907\+00:00: rank \#40

2026\-10\-09T21:02:08\.376144\+00:00: rank \#42

2026\-10\-09T22:01:19\.253075\+00:00: rank \#44

2026\-10\-09T23:01:31\.990912\+00:00: rank \#44

2026\-10\-10T08:01:00\.593376\+00:00: rank \#43

2026\-10\-10T09:01:45\.527975\+00:00: rank \#45

2026\-10\-10T10:00:53\.799353\+00:00: rank \#45

2026\-10\-10T11:00:35\.65851\+00:00: rank \#44

2026\-10\-10T12:01:11\.169647\+00:00: rank \#44

2026\-10\-10T13:00:54\.008377\+00:00: rank \#42

2026\-10\-10T14:01:43\.208766\+00:00: rank \#44

2026\-10\-10T15:01:24\.825188\+00:00: rank \#44

2026\-10\-10T16:01:10\.213328\+00:00: rank \#43

2026\-10\-10T17:00:48\.84189\+00:00: rank \#42

2026\-10\-10T18:00:57\.55132\+00:00: rank \#42

2026\-10\-10T19:00:52\.379463\+00:00: rank \#41

2026\-10\-10T20:00:21\.993912\+00:00: rank \#41

2026\-10\-10T21:01:10\.657738\+00:00: rank \#41

2026\-10\-10T22:00:50\.433569\+00:00: rank \#40

2026\-10\-10T23:01:00\.95027\+00:00: rank \#41

Cloudflare's Clef models claim faster, typed decisions than Jev, but the only supplied comment asks how such models were replicated so quickly, leaving novelty and copyability unresolved\.

## The brief

Cloudflare has released Clef and Clef\-flash, two decision models hosted on Workers AI and open\-sourced under Apache 2\.0\. The models classify inputs into typed, probabilistic outputs rather than generating text, using a Qwen backbone with a non\-autoregressive scoring pass\. Cloudflare says Clef leads the Jev Decision Index and reports lower latency than Jev and gpt\-oss\-120b in an internal domain\-classification workflow, while also launching an RL fine\-tuning service\.

- Clef uses a vision encoder and 64k context window; Cloudflare says Jev is text\-only with 32k context\.
- In a Threat Intelligence test, Clef fetched, rendered, and classified a domain in 2\.2s versus 4\.7s for gpt\-oss\-120b, which returned two classifications\.
- Cloudflare reports Clef and Clef\-flash beat Jev on latency across 43 evals, except Laya, which is faster but lower quality on listed benchmarks\.
- Training freezes Qwen3\.8\-27B for Clef and Qwen3\.5\-9B for Clef\-flash, optimizing a routing head and rank\-256 LoRA adapters with label\-smoothed cross\-entropy and Brier loss\.
- The new RL service uses AI Gateway, Workers AI, Containers, a Trainer, and BYO Model to fine\-tune and redeploy Clef\.
- Cloudflare says it does not read, store, or train on customer requests or responses unless customers use fine\-tuning\.

## Discussion themes

Analyzed: 2026\-10\-02T16:02:40\.295527\+00:00

Analysis sample: Based on 24 of 24 usable stored comments. Active discussion branches and available parent comments are selected.

This sample may omit parts of the full thread. Selected themes do not measure community opinion or how common a view is.

### Reproducibility and benchmark validity concerns

Commenters challenge the open\-source framing because training data and pipeline are not published, and question whether Clef's Jev Decision Index lead holds up: one reports slower and worse moderation performance than Jev, another says many models claiming Jev\-level performance fail on non\-trivial tasks\.

Sources: [Comment 49924502](<https://news.ycombinator.com/item?id=49924502>) · [Comment 49929048](<https://news.ycombinator.com/item?id=49929048>) · [Comment 49927360](<https://news.ycombinator.com/item?id=49927360>) · [Comment 49924445](<https://news.ycombinator.com/item?id=49924445>)

### Clef pricing and self\-hosting economics

Commenters compare Jev and Clef input pricing, estimate per\-million\-decision costs, and suggest self\-hosting Clef if resources allow; one notes the higher price but still considers trying it because of the embedded vision encoding\.

Sources: [Comment 49926385](<https://news.ycombinator.com/item?id=49926385>) · [Comment 49924297](<https://news.ycombinator.com/item?id=49924297>)

### Vision encoder performance and accuracy

One commenter values the embedded vision encoding enough to try Clef despite higher pricing; another tests an 8\-bit quant on coin classification and reports poor accuracy and slower speed than a local Gemma model\.

Sources: [Comment 49933256](<https://news.ycombinator.com/item?id=49933256>) · [Comment 49924297](<https://news.ycombinator.com/item?id=49924297>)

### Local and smaller\-model alternatives

Commenters ask about running Jev\-like models locally or via cheap APIs, and promote a 60M local model (At0M) with low latency and decent accuracy as a smaller alternative; one also describes making a basic decision model in minutes from existing open\-source models\.

Sources: [Comment 49925083](<https://news.ycombinator.com/item?id=49925083>) · [Comment 49925524](<https://news.ycombinator.com/item?id=49925524>) · [Comment 49925771](<https://news.ycombinator.com/item?id=49925771>) · [Comment 49924401](<https://news.ycombinator.com/item?id=49924401>)

### Fine\-tuning, priors, and customization

One commenter argues Jev\-like models are not useful unless they can be fine\-tuned or accept priors, saying the Jev API handles priors poorly; another notes Clef supports instruction and RL fine\-tuning, though RL tuning may be limited to Cloudflare customers\.

Sources: [Comment 49925842](<https://news.ycombinator.com/item?id=49925842>) · [Comment 49933702](<https://news.ycombinator.com/item?id=49933702>)

### Data privacy concerns with hosted Cloudflare models

A commenter questions why users would give their data to Cloudflare and promotes a local 60M alternative instead\.

Sources: [Comment 49925524](<https://news.ycombinator.com/item?id=49925524>)

## Sources & coverage

AI-generated summary · 2026\-10\-01T18:01:21\.047874\+00:00

Based on 1 of 1 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. Article text may also be shortened.

Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.