Hey can you share an example payload you’re sending for polling the batch request

KKevin Hey can you share an example payload you’re sending for polling the batch reques...

A

Ashkan•9/29/25, 10:17 PM

I've had that error almost days ago(I can't remember the payload) but the problem was related to JSON format of the

payload

payload

payload

payload and I could resolve the error. although the error resolved and no more exceptions were thrown but when I was checking status of the batch job it was showing

invalid parameters

invalid parameters

invalid parameters

invalid parameters. so I decided to suspend my works around the batch processing until stable version of that is announced.

AAshkan I've had that error almost days ago(I can't remember the payload) but the proble...

A

Ashkan•9/29/25, 10:22 PM

this is the part of the code I was using for batch process:

const response = args.env.AI.run(
                    '@cf/openai/gpt-oss-120b' as any,
                    {
                        requests: [input],
                    },
                    { queueRequest: true, gateway: { id: GATEWAY_ID } }
                );

const response = args.env.AI.run(
                    '@cf/openai/gpt-oss-120b' as any,
                    {
                        requests: [input],
                    },
                    { queueRequest: true, gateway: { id: GATEWAY_ID } }
                );

const response = args.env.AI.run(
                    '@cf/openai/gpt-oss-120b' as any,
                    {
                        requests: [input],
                    },
                    { queueRequest: true, gateway: { id: GATEWAY_ID } }
                );

const response = args.env.AI.run(
                    '@cf/openai/gpt-oss-120b' as any,
                    {
                        requests: [input],
                    },
                    { queueRequest: true, gateway: { id: GATEWAY_ID } }
                );

A

Ashkan•9/29/25, 10:24 PM

it should be noted

input

input

of type

Array

Array

worked but I wasn't able to run it when

input

input

was of type

string

string

(openai compatible JSON format)

D

displayname•9/29/25, 10:44 PM

Hey, could we please get some eyes on workers-and-pages-helpOpenAI Completions compatibility for gpt-oss models?
Sorry I didn't know where to open the thread

A

ajgeiss0702•9/30/25, 1:23 AM

just got an error i havent seen before with

gpt-oss-20b

gpt-oss-20b

gpt-oss-20b

gpt-oss-20b

AiError: AiError: EngineCore encountered an issue. See stack trace (above) for the root cause. (50d9f590-dc24-4cf3-92b4-313011575443)

AiError: AiError: EngineCore encountered an issue. See stack trace (above) for the root cause. (50d9f590-dc24-4cf3-92b4-313011575443)

AiError: AiError: EngineCore encountered an issue. See stack trace (above) for the root cause. (50d9f590-dc24-4cf3-92b4-313011575443)

AiError: AiError: EngineCore encountered an issue. See stack trace (above) for the root cause. (50d9f590-dc24-4cf3-92b4-313011575443)

account id:

f55b85c8a963663b11036975203c63c0

f55b85c8a963663b11036975203c63c0

f55b85c8a963663b11036975203c63c0

f55b85c8a963663b11036975203c63c0
(there is no stack trace above)

Z

Zetax•10/1/25, 6:29 AM

I wish the types would work for this

Wanted to try out a model and the response type for it is completely broken.

Bildschirmfoto_2025-10-01_um_08.28.55.png

J

James•10/2/25, 6:34 PM

Workers AI types need a lot of love, yeah.
- https://github.com/cloudflare/workerd/issues/2181
- https://github.com/cloudflare/workerd/issues/2418
- https://github.com/cloudflare/workerd/issues/5080

N

Nova Sven Cat•10/3/25, 7:14 PM

I'm using workers AI with the free plan and it still works even when I reached the quota, what is happening?

N

Nova Sven Cat•10/3/25, 8:00 PM

Daily usage (resets at 00:00 UTC) - Neurons used today: 32.48k/10k

N

Nova Sven Cat•10/3/25, 8:28 PM

You have exceeded the Neuron limit of 10,000 for this today. Upgrade your plan

NNova Sven Cat > You have exceeded the Neuron limit of 10,000 for this today. Upgrade your plan

Z

Zetax•10/3/25, 9:58 PM

why would you report that though

N

Nova Sven Cat•10/3/25, 9:58 PM

i have no idea why it's still working so i thought it was a bug

Z

Zetax•10/3/25, 9:58 PM

Maybe it's intentional, some overage to not discourage customers

Z

Zetax•10/3/25, 9:59 PM

Not sure

Z

Zetax•10/3/25, 9:59 PM

Although 3x seems a bit much lol

N

Nova Sven Cat•10/3/25, 9:59 PM

yeah im using the ai for an irc bot lmao

ZZetax Maybe it's intentional, some overage to not discourage customers

N

Nova Sven Cat•10/3/25, 10:02 PM

from the docs it says any future reqs will error if I reach that neuron quota :P

All limits reset daily at 00:00 UTC. If you exceed any one of the above limits, further operations will fail with an error.

A

Anon•10/4/25, 1:41 AM

Hey! I'm a software dev. I'm interested in building a RAG system and would like to hear y'all opinions on this platform when it comes to Workers AI and Vectorize. Thanks :3

A

Ashkan•10/5/25, 12:02 AM

Hey, how can I have message history such that ai model will have a message history and depend on this history will answer. for example.

User: sum of 2 + 2? -> Model: the answer is 4 -> User:now multiply the result by 2

User: sum of 2 + 2? -> Model: the answer is 4 -> User:now multiply the result by 2

at last user user input, ai model should have a history of messages and know what the

result

result

is.

AAshkan Hey, how can I have message history such that ai model will have a message histo...

I

Isaac McFadyen•10/5/25, 12:07 AM

That is possible but Workers AI will not do that for you. You need to store the messages somewhere (or have the client re-send the full history) and then send the full conversation with each Workers AI request.

IIsaac McFadyen That is possible but Workers AI will not do that for you. You need to store the ...

A

Ashkan•10/5/25, 12:11 AM

thanks for reply, this approach may result in exponential

token_input

token_input

and exponential cost

AAshkan thanks for reply, this approach may result in exponential `token_input` and expo...

I

Isaac McFadyen•10/5/25, 12:13 AM

That is just how LLMs work and how every provider does it, not just Workers AI. OpenAI, Gemini, Claude, etc all need to see the past history in order to act on it. If a message isn't in the context then it will not be seen by the model.

IIsaac McFadyen That is just how LLMs work and how every provider does it, not just Workers AI. ...

A

Ashkan•10/5/25, 12:14 AM

ok, I will try it. thanks for help

AAshkan Hey, how can I have message history such that ai model will have a message histo...

A

Anon•10/5/25, 2:16 AM

Use Durable Obejcts to store the history

A

Anon•10/5/25, 2:17 AM

At least that's what's recommended you could really store them anywhere

C

Comrade Nalaxone•10/5/25, 4:44 PM

For god sake can you all please update the docs to match the portal this drives me nuts every single time:

https://developers.cloudflare.com/workers-ai/get-started/rest-api/#1-get-api-token-and-account-id

Cloudflare Docs

REST API

Use the Cloudflare Workers AI REST API to deploy a large language model (LLM).

C

Comrade Nalaxone•10/5/25, 4:44 PM

Too big of a company for that to happen

L

LordSilver•10/6/25, 9:21 PM

what is the default model being used?

LLordSilver what is the default model being used?

I

Isaac McFadyen•10/6/25, 9:32 PM

There is no default, you pick whatever model on Workers AI you want to use yourself.

IIsaac McFadyen There is no default, you pick whatever model on Workers AI you want to use yours...

L

LordSilver•10/6/25, 9:37 PM

but i didn't pick anything

I

Isaac McFadyen•10/6/25, 9:38 PM

If you're using Workers AI you must have? Are you using another product of some sort?

L

LordSilver•10/6/25, 9:38 PM

ok it uses llama by default

IIsaac McFadyen If you're using Workers AI you must have? Are you using another product of some ...

L

LordSilver•10/6/25, 9:38 PM

i use Workes AI and i don't specify any model

LLordSilver i use Workes AI and i don't specify any model

I

Isaac McFadyen•10/6/25, 9:39 PM

You must be using a product that uses Workers AI and picks the model for you, there are no default models in Workers AI. I saw in #workers-and-pages-discussions that you're using Stagehand, correct? I expect that picks llama-3.3 for you.

L

LordSilver•10/6/25, 9:40 PM

mmm ok

S

sodepr•10/9/25, 6:15 AM

Hi,

Does Cloudflare support legacy endpoints (

/completions

/completions

,

/chat/completions

/chat/completions

) for new models (Eg: gpt-oss-20b)?

The document said that:

Workers AI supports OpenAI compatible endpoints for text generation (/v1/chat/completions)

Workers AI supports OpenAI compatible endpoints for text generation (/v1/chat/completions)

But I cannot use any new model.

A

Ashkan•10/9/25, 9:49 AM

doesn't

runWithTools

runWithTools

support

gateway

gateway

configuration?

Ssodepr Hi, Does Cloudflare support legacy endpoints (`/completions`, `/chat/completion...

K

Kevin•10/9/25, 9:52 AM

Use responses endpoint.

KKevin Use responses endpoint.

S

sodepr•10/9/25, 2:46 PM

Applications hasn't support new endpoint yet.

Ssodepr Applications hasn't support new endpoint yet.

K

Kevin•10/9/25, 2:51 PM

at least ai gateway does support it ai-gateway I have not tried workers ai directly

H

Hew•10/9/25, 7:27 PM

@sodepr The Cloudflare blog said in August that completions support was "coming soon" to the gpt-oss models. I haven't seen any update from them since, though. It would certainly be nice to have because, as you've pointed out, not everything supports the responses endpoint yet.

R

Razmjoo•10/9/25, 8:44 PM

Hi, is there any way to use

@cf/tomarkdown

@cf/tomarkdown

/

@cf/markdown

@cf/markdown

using the AI Gateway?

M

Manqo•10/10/25, 11:27 PM

Hello! I Been looking at worker ai For txt2img. Im quite new with these different Image models, i Have usually just used Gemini or chatgpt Image generation. But to cut down on spending im intrested of using worker ai to generate images. So My question which of The avaivable Image models IS best to create Hand draw pictures? I dont need realism. I tried Flux schell from huggingface but that tend to not create images that Look like drawings.

C

crazyjack12•10/11/25, 11:19 PM

Maybe stable-diffusion-xl-base-1.0

S

Sudhan•10/12/25, 11:56 AM

hey guys i have question. Lets say i have an worker endpoint for processing a audio file. The audio file is 500mb
The endpoint recieves the url and downloads during the endpoint runtime, and then transcribe the audio to text and then vectorize it.. I mean it takes a lot of time cpu compute so is it possible in workers?

MManqo Hello! I Been looking at worker ai For txt2img. Im quite new with these differen...

S

samjs•10/12/25, 6:02 PM

You could try lucid-origin? It generates some nice images in my experience. e.g. here's a short example:

lucidOrigin.get('/', async (c) => {
    // Generate the image
    const response = await c.env.AI.run(
        "@cf/leonardo/lucid-origin",
        {
            prompt: "picture in a hand-drawn style of a mini goldendoodle dog playing fetch in a field"
        }
    );
    console.log(response);

    // Embed the image directly as a data URL
    const imageDataUrl = `data:image/jpeg;base64,${response.image}`;

    const html = `
<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>AI Image Generator</title>
    <style>
        // styles
    </style>
</head>
<body>
    <div class="container">
        <h1>🎨 Leonardo AI Image Generator</h1>
        <p>Generating: <em>A mini goldendoodle dog playing fetch in a field (hand-drawn style)</em></p>
        <div class="image-container">
            <img src="${imageDataUrl}" alt="AI Generated Image" />
        </div>
    </div>
</body>
</html>
    `;
    return c.html(html);
});

lucidOrigin.get('/', async (c) => {
    // Generate the image
    const response = await c.env.AI.run(
        "@cf/leonardo/lucid-origin",
        {
            prompt: "picture in a hand-drawn style of a mini goldendoodle dog playing fetch in a field"
        }
    );
    console.log(response);

    // Embed the image directly as a data URL
    const imageDataUrl = `data:image/jpeg;base64,${response.image}`;

    const html = `
<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>AI Image Generator</title>
    <style>
        // styles
    </style>
</head>
<body>
    <div class="container">
        <h1>🎨 Leonardo AI Image Generator</h1>
        <p>Generating: <em>A mini goldendoodle dog playing fetch in a field (hand-drawn style)</em></p>
        <div class="image-container">
            <img src="${imageDataUrl}" alt="AI Generated Image" />
        </div>
    </div>
</body>
</html>
    `;
    return c.html(html);
});

lucidOrigin.get('/', async (c) => {
    // Generate the image
    const response = await c.env.AI.run(
        "@cf/leonardo/lucid-origin",
        {
            prompt: "picture in a hand-drawn style of a mini goldendoodle dog playing fetch in a field"
        }
    );
    console.log(response);

    // Embed the image directly as a data URL
    const imageDataUrl = `data:image/jpeg;base64,${response.image}`;

    const html = `
<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>AI Image Generator</title>
    <style>
        // styles
    </style>
</head>
<body>
    <div class="container">
        <h1>🎨 Leonardo AI Image Generator</h1>
        <p>Generating: <em>A mini goldendoodle dog playing fetch in a field (hand-drawn style)</em></p>
        <div class="image-container">
            <img src="${imageDataUrl}" alt="AI Generated Image" />
        </div>
    </div>
</body>
</html>
    `;
    return c.html(html);
});

lucidOrigin.get('/', async (c) => {
    // Generate the image
    const response = await c.env.AI.run(
        "@cf/leonardo/lucid-origin",
        {
            prompt: "picture in a hand-drawn style of a mini goldendoodle dog playing fetch in a field"
        }
    );
    console.log(response);

    // Embed the image directly as a data URL
    const imageDataUrl = `data:image/jpeg;base64,${response.image}`;

    const html = `
<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>AI Image Generator</title>
    <style>
        // styles
    </style>
</head>
<body>
    <div class="container">
        <h1>🎨 Leonardo AI Image Generator</h1>
        <p>Generating: <em>A mini goldendoodle dog playing fetch in a field (hand-drawn style)</em></p>
        <div class="image-container">
            <img src="${imageDataUrl}" alt="AI Generated Image" />
        </div>
    </div>
</body>
</html>
    `;
    return c.html(html);
});

SSudhan hey guys i have question. Lets say i have an worker endpoint for processing a au...

S

samjs•10/12/25, 6:07 PM

Hey @Sudhan . There are a couple of things that might help here:

1. you can pass in a stream of bytes for the inputs in many of the models, e.g. in the nova-3 example you can fetch and pass in the body directly (without needing to await it: https://developers.cloudflare.com/workers-ai/models/nova-3/)
2. Some models (like nova-3) support our async batch API (https://developers.cloudflare.com/workers-ai/features/batch-api/). So you can submit the job for running the inference and then poll for the response.

In both cases, it shouldn't be using much compute in the worker itself.

J

John Spurlock•10/12/25, 11:10 PM

Does anyone know what the

"p"

"p"

property represents when streaming results back from

@cf/google/gemma-3-12b-it

@cf/google/gemma-3-12b-it

? I can't find any docs on it

P

Peps•10/12/25, 11:28 PM

its padding

JJohn Spurlock Does anyone know what the `"p"` property represents when streaming results back ...

P

Peps•10/12/25, 11:28 PM

the reason for the padding is more covered here: https://blog.cloudflare.com/ai-side-channel-attack-mitigated/

The Cloudflare Blog

Mitigating a token-length side-channel attack in our AI products

The Workers AI and AI Gateway team recently collaborated closely with security researchers at Ben Gurion University regarding a report submitted through our Public Bug Bounty program. Through this process, we discovered and fully patched a vulnerability affecting all LLM providers. Here’s the story

Hey can you share an example payload you’re sending for polling the batch request

Similar Threads

Similar Threads

Similar Threads