arcprize/arc-agi-3-benchmarking

55 starsPython Latest
Open on GitHub

What It Calls

Calls external APIs Your API key Paid Free tier The repository's own service

Without a key: Text generation · Image generation · Speech-to-text · Text-to-speech · Embeddings (OpenAI): has a branch that stops with an error without a key.

Feature matrix
Feature Provider Use Default Alternative
Text generation · Image generation · Embeddings Google Gemini API
6 call sites
Key needed generativelanguage.googleapis.com Other providers
Text generation · Image generation · Speech-to-text · Text-to-speech · Embeddings OpenAI
1 call sites
Key needed api.openai.com None
Text generation anthropic, gemini, openrouter, xai
4 call sites
Key needed api.x.ai, openrouter.ai +2 more None
External API (arcprize.org) arcprize.org The repository's own service
1 call sites
Undetermined arcprize.org None
Text generation Anthropic Claude
1 call sites
Optional api.anthropic.com Other providers
Text generation OpenRouter
1 call sites
Optional openrouter.ai Other providers
Text generation · Image generation xAI Grok
2 call sites
Optional api.x.ai Other providers
External APIs
API Credential Code evidence Pricing Pricing page Checked
Google Gemini API Your API key benchmarking/runtime_adapters.py:413Text generation, image generation, and embeddings are provided via the Google Gemini API, which requires a Google Gemini API key. Free tier Pricing 2026-10-05
OpenAI Your API key benchmarking/chat_streaming.py:71Text and image generation, speech-to-text, text-to-speech, and embeddings are provided via OpenAI; if the call fails, it results in an error. Paid Pricing 2026-10-05
arcprize.orgThe repository's own service Credential unclear main.py:60Not shown to stop without a key or address No pricing info - -
Anthropic Claude Your API key benchmarking/runtime_clients.py:67Anthropic Claude can be used for text generation, though other providers may be chosen instead. Paid Pricing 2026-10-05
OpenRouter Your API key benchmarking/runtime_clients.py:56OpenRouter can be used for text generation, though other providers may be chosen instead. Free tier Pricing 2026-10-05
xAI Grok Your API key benchmarking/runtime_clients.py:54xAI Grok can be used for text and image generation, though other providers may be chosen instead. Paid Pricing 2026-10-05

Data sent out: prompt text · 5 provider(s) unconfirmed

Checked scope: Full check — all 61 code and config files (76 files in the repository) · tree not truncated · no submodules

Next

Check another repository This result is wrong

Badge preview

external API badge

[![external API](https://whatitcalls.reactiveworks.dev/badge/arcprize/arc-agi-3-benchmarking.svg)](https://whatitcalls.reactiveworks.dev/en/arcprize/arc-agi-3-benchmarking)

Shows as “external API · used” in a README.

Was this analysis helpful?

Freshness detailscommit 3239e22 · checked 2026-10-07
Committed
2026-10-06 21:37 UTC
Checked
2026-10-07 11:48 UTC
HEAD last seen
2026-10-07 11:47 UTC · 3239e22 (same as the checked commit)
Read depth
Basic (up to 400 files · 8 MB)
Scanner · rules
scan-5+verdict-4
Pricing catalog
version 2
Correction revision
0
Summary
Model ollama-cloud/gemma4:31b-cloud · some rule sentences
What is sent out
README claims ↔ code

No README claim contradicts the code.

Code quotes

Source files are not stored. Code is linked at the checked commit on GitHub; stored quotes are at most 3 lines per evidence and 6 per file.

Source: arcprize/arc-agi-3-benchmarking · commit 3239e22 · benchmarking/runtime_clients.py

  • L11 host
  • L54 SDK call
  • L56 host
  • L67 SDK call
  • 1 more locations

Source: arcprize/arc-agi-3-benchmarking · commit 3239e22 · benchmarking/runtime_adapters.py

Source: arcprize/arc-agi-3-benchmarking · commit 3239e22 · benchmarking/chat_streaming.py

Checked scope details

Files

In the repository
76
Checked
61 (80%)

Manifests

All priority files read

  • pyproject.toml

Unresolved import paths (0)

None

Excluded files

  • .md 9
  • (none) 3
  • .ini 1
  • .yaml 1
  • lockfile 1
External code not checked
None
Limits hit
None
Correction history

No corrections yet.

This result is wrong · Corrections policy