Hawk or dove?

# Jev vs Chat Model: Hawk or Dove?

**Live:** https://fedspeech.pages.dev

Real-time analysis of Fed press conferences with TypeSafe's **Jev**, side by side with a chat model. Is the Chair hawkish or dovish? Jev scores **every word** the Chair says in about 150 ms; the chat model scores each caption line in a few seconds. Both are judged against the same four stances: Hawkish, Partly hawkish, Partly dovish and Dovish.

Three ways to try it:

- **Watch the demo:** a recorded run of the July 29, 2026 press conference (Chairman Warsh), replayed with real answers and real timings. It makes no API calls.
- **Past Fed press conference:** pick a date. The page plays the Fed's official video with the Fed's official transcript as captions and judges only the Chair, not reporters. Answers are saved, so each conference is paid for once and replays free after that. Switch the chat model mid-video without reloading.
- **Type or paste:** judge any statement as you type.

The dating edition (Red Flag Detector, Is She/He Interested?) is in [TurboGuo/jev-dating](https://github.com/TurboGuo/jev-dating).

## Chat models

| Model | On the site's account (Cloudflare AI Gateway) | With a visitor's own key |
|---|---|---|
| Claude Opus 5 | `anthropic/claude-opus-5`, effort low | Anthropic API |
| GPT-5.6 Sol | `openai/gpt-5.6-sol`, reasoning low | OpenAI API |
| GLM-5.3 | `@cf/zai-org/glm-5.3`, reasoning low | Z.ai API |
| DeepSeek V4 Flash | `@cf/deepseek-ai/deepseek-v4-flash-0731`, reasoning low | DeepSeek API |

Models are listed in `src/models.ts`; set `enabled: false` to turn one off.

## How it works

One Cloudflare Worker: `public/` is the page and `src/` is the API. A Cloudflare Pages site (`pages/`) is the public address and forwards every request to the Worker over a service binding