A new bird in stealth

Big bird.
Serious code.

An anonymous coding model with a 262K context window and strong Next.js results.
Bring a little patience.

Try Pixel Canary See the results

Say hello.

Next.js tasks · AGENTS.md: 30/31
Context window: 262K tokens
Temporary preview: $0 token price

Next.js Agent Evals

How Pixel Canary
compares.

With AGENTS.md, Pixel Canary passed 30 of 31 tasks—matching Opus, Astra, Gemini and Kimi in this evaluation.

The table compares model-and-agent combinations on the same Next.js task suite.

What does AGENTS.md change?

31 tasks. Familiar company.

Next.js task success

Task success by model and agent, with baseline and AGENTS.md scores shown together.
Model / coding agent	Baseline	With `AGENTS.md`
Pixel CanaryOpenCode	90%28/31	97%30/31
Claude Opus 5.5 (high)Claude Code	97%30/31	97%30/31
GPT 6 Astra (high)Codex	90%28/31	97%30/31
Gemini 3.8 FlashOpenCode	90%28/31	97%30/31
Grok 4.7OpenCode	94%29/31	94%29/31
Kimi K3OpenCode	84%26/31	97%30/31

September 25, 2026 snapshot · pass@4: a task passes if any of up to four attempts succeeds. Full leaderboard ↗

Give the bird directions

What does AGENTS.md change?

It adds project instructions telling the coding agent to read the relevant Next.js documentation before coding.

Both runs have the docs in the installed Next.js package. The added AGENTS.md file points the agent to them.

See the exact instructions

Two more. Nice.

Tap the bird to compare ↔

30/31tasks passed

PassedGained with AGENTS.mdNot passed

31 tasks · OpenCode · one tile per task

A heavyweight. At its own pace.16.9 min average evaluation duration

An average across both test conditions. A task passes if any of up to four attempts succeeds.

See how it was tested

The two tasks AGENTS.md helped solve

038 · refresh-settingsFail → Pass

048 · prefetch-url-dataFail → Pass

044 · uses-nextjsFail → Fail

Two additional successes: 90.32% becomes 96.77%. The remaining task asks for a book-tracking app with JSON-file persistence.

The harness uses Gateway-default reasoning, up to four attempts with early exit, and a 40-minute timeout per attempt. The 1,015.8-second average is evaluation duration, not time to first token or the complete suite runtime.

Pixel Canary configuration ↗Published results ↗

Give your characters a voice.

Expressive dialogue and original music, created on your own Windows PC with Demodokos Foundry.

Explore Demodokos Start making voices

Ooh. Story time.

Tap the bird to listen ↘

A little listening break.

A DEMODOKOS DEMO

Barnaby Bear

Barnaby Bear · three-voice Demodokos demo

Press play. Meet Barnaby.

Voices with character.Demodokos

Take the bird for a spin

Put Pixel Canary
to work.

$0 / tokenInput & output · temporary stealth preview

Try a prompt in your browser, choose it in your coding client, or make your first call through Vercel AI Gateway.

Model IDstealth/pixel-canary

Vercel AI GatewayOfficial listing · account & API key AI SDK PlaygroundStart with a prompt in your browser Command CodeGo plan and above · subscription separate

Also in Cline’s free preview. Using OpenCode? See Vercel’s evaluation configuration.

Preview data policy The provider may retain prompts and outputs for training. Read the listing

One pixel at a time.

Tap for the model ID ↗

Chat completions / cURL

: "${AI_GATEWAY_API_KEY:?Set AI_GATEWAY_API_KEY first}"

curl --fail-with-body --silent --show-error \
  --max-time 180 \
  https://ai-gateway.vercel.sh/v1/chat/completions \
  -H "Authorization: Bearer ${AI_GATEWAY_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "stealth/pixel-canary",
    "messages": [{
      "role": "user",
      "content": "Write a TypeScript function that validates SRT timestamp ordering. Include three unit tests."
    }],
    "max_tokens": 4096,
    "stream": false
  }'

Set AI_GATEWAY_API_KEY in your shell. The 180-second timeout is a client setting in this example.

Gateway API documentation

Room for the whole conversation.

Prompt and response share the context window.

Context window: 262,144 tokens
Maximum output: 131,072 tokens
Input → output: Text + image → text

Reasoning, caching & tool use

The catalog offers four reasoning settings: none, low, medium and xhigh, plus implicit caching. The output limit sits within the total context allowance.

Vercel’s OpenCode evaluation enables tool use in its model configuration.

Open the model catalog

The name is out. The maker isn’t.

One sizeable
little mystery.

Who, me?

Pixel CanaryIn stealth since September 25, 2026

Pixel Canary’s developer
remains undisclosed.

Two familiar words make Google an interesting guess. Follow the name.

A little detective work.

01 · The first word
Pixel. A familiar name.
Google lists Pixel™ among its trademarks for phones, tablets, smartwatches and accessories.
Google’s trademark list
02 · The second word
Canary. Another connection.
Google also ships Chrome Canary, the experimental version of its browser.
Meet Chrome Canary
03 · The missing piece
A theory. Still no reveal.
Those name matches suggest Google. Vercel’s model listing still identifies the provider only as “stealth”.
Check the official listing

See the community guesses

A few useful answers

Before you
say hello.

Quality, pace and the details of the preview.

Try Pixel Canary

Is Pixel Canary good at coding?

Its strongest published result is Next.js development: 28 of 31 tasks passed in the baseline, and 30 of 31 with an AGENTS.md instruction to consult the bundled Next.js docs, using OpenCode and pass@4. Explore the results.

Is Pixel Canary slow?

Vercel’s Next.js evaluation reports a 16.9-minute average duration across both conditions. Early OpenCode users also describe long generation and code-reading waits. Those reports are separate from the timed evaluation. Read their experiences.

Is Pixel Canary free?

Vercel lists $0 input and output token prices during the temporary stealth preview. Coding clients can have separate subscriptions; Command Code requires its Go plan or above. See the access options.

Who made Pixel Canary?

The developer remains undisclosed as of September 26, 2026. The “stealth” label on Vercel does not identify the company. More about the identity.

Can Pixel Canary take images?

Yes. Vercel’s catalog lists text and image input, with text output. The context window is 262,144 tokens; generated output is capped at 131,072 tokens within that total. View the catalog.

Can prompts be used for training?

Yes. The preview listing says the provider may retain prompts and outputs for training. Read the current data policy.

Straight to the sources

The details
behind the bird.

Checked September 26, 2026.
Benchmark snapshot: September 25.

01
Vercel launch announcement
Launch, free preview and initial results
02
Vercel model listing
Specifications, access and data policy · API catalog
03
Next.js Agent Evals
Scores, duration and methodology
04
Open evaluation repository
Harness and task definitions · Results JSON
05
Command Code model guide
Client access and plan requirement
06
Community testing discussion
First impressions and identity theories

Big bird. Serious code.

How Pixel Canary compares.

Next.js task success

What does AGENTS.md change?

Give your characters a voice.

Barnaby Bear

Put Pixel Canary to work.

Room for the whole conversation.

One sizeable little mystery.

A little detective work.

Pixel. A familiar name.

Canary. Another connection.

A theory. Still no reveal.

Before you say hello.

The details behind the bird.

Big bird.
Serious code.

How Pixel Canary
compares.

Put Pixel Canary
to work.

One sizeable
little mystery.

Before you
say hello.

The details
behind the bird.