Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion apps/web/content/blog.md
Original file line number Diff line number Diff line change
Expand Up @@ -93,9 +93,11 @@ Measured response times in dev through AI Gateway:
| Scenario | Model | Time |
| :------------------------------ | :------------ | -----: |
| Single-sentence answer | Llama 3.3 70B | 5.1 s |
| Two attachments with reasoning | Qwen3 30B | 26.3 s |
| Two attachments with reasoning | Qwen3.8 27B | 26.3 s |
| Short summary (109 in / 42 out) | Llama 3.3 70B | 1.7 s |

The 26.3 s figure was measured on `@cf/qwen/qwen3-30b-a3b-fp8` before Workers AI retired that model on 20 November 2026. The picker now calls `@cf/qwen/qwen3.8-27b`.

## Quick start

```sh
Expand Down
4 changes: 3 additions & 1 deletion apps/web/content/blog.zh.md
Original file line number Diff line number Diff line change
Expand Up @@ -93,9 +93,11 @@ await ai.stream(model, { messages, max_tokens: 2048 }, { gateway: { id: "void-st
| 测试场景 | 模型 | 耗时 |
| :----------------------------- | :------------ | -----: |
| 单句回答 | Llama 3.3 70B | 5.1 s |
| 两个附件 + 深度思考 | Qwen3 30B | 26.3 s |
| 两个附件 + 深度思考 | Qwen3.8 27B | 26.3 s |
| 简短摘要(109 输入 / 42 输出) | Llama 3.3 70B | 1.7 s |

上表 26.3 s 测自 `@cf/qwen/qwen3-30b-a3b-fp8`。Workers AI 于 2026 年 11 月 20 日停用该模型,模型选择器现已改用 `@cf/qwen/qwen3.8-27b`。

## 快速开始

```sh
Expand Down
4 changes: 2 additions & 2 deletions apps/web/src/lib/models.ts
Original file line number Diff line number Diff line change
Expand Up @@ -10,11 +10,11 @@ export interface Model {

/**
* Workers AI models the composer offers. They run on Cloudflare's free daily allowance, so the
* app needs no provider key. Qwen3 and gpt-oss stream their reasoning before the answer.
* app needs no provider key. Qwen3.8 and gpt-oss stream their reasoning before the answer.
*/
export const MODELS = [
{ id: "@cf/meta/llama-3.3-70b-instruct-fp8-fast", name: "Llama 3.3 70B", provider: "meta" },
{ id: "@cf/qwen/qwen3-30b-a3b-fp8", name: "Qwen3 30B", provider: "qwen" },
{ id: "@cf/qwen/qwen3.8-27b", name: "Qwen3.8 27B", provider: "qwen" },
{ id: "@cf/openai/gpt-oss-20b", name: "gpt-oss 20B", provider: "openai" },
{
id: "@cf/mistralai/mistral-small-3.1-24b-instruct",
Expand Down
2 changes: 2 additions & 0 deletions docs/deployment.md
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,8 @@ Workers AI calls name the gateway on every request (`AI_GATEWAY_ID` in `apps/web

The model picker lists `@cf/*` models from `apps/web/src/lib/models.ts`. They need no provider key: usage counts against the account's Workers AI allowance, measured in neurons. Some models need access the account may not have; `@cf/google/gemma-3-12b-it`, for one, answered `403` ("This account is not allowed to access") on the account this starter was built with. Try a model in dev before you add it.

Workers AI retires `@cf/qwen/qwen3-30b-a3b-fp8` on 20 November 2026. The picker offers `@cf/qwen/qwen3.8-27b` (Qwen3.8 27B) in its place.

### Usage limits

There are no accounts, so once the `workers.dev` route is on, anyone with the URL can send messages. On the Workers Free plan, Workers AI stops at the daily free allowance and the chat answers `429`. On Workers Paid, usage past the allowance is billed. Set a rate limit on the gateway (AI Gateway › Settings › Rate limiting) before you share the URL widely.
Expand Down
Loading