# ARC-AGI-3 Semi-Private: Provider Adapter harness (ARC-AGI-3 Provider Adapter)

> ARC Prize verified interactive reasoning results using a provider-specific stateful adapter, kept separate from Standard harness results.

Canonical page: https://benchlm.ai/benchmarks/arc-agi-3-provider-adapter

- Category: [Reasoning](/reasoning)
- Last updated: Captured 2026-10-28

## About ARC-AGI-3 Provider Adapter

- Year: 2026
- Paper: [ARC Prize verified ARC-AGI-3 results](https://arcprize.org/leaderboard)

The Provider Adapter preserves opaque reasoning state between requests and compacts long conversations. The table retains each published effort configuration. Model pages select the highest declared published effort within this harness, while the table also preserves lower-effort results such as Gemini 4 Argon at 97.459172% with low reasoning.

ARC-AGI-3 Provider Adapter is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (43 configurations)

| Rank | Model | Configuration | Creator | Score |
|------|-------|---------------|---------|-------|
| 1 | [GPT-6 Astra - Provider Adapter (High)](/models/gpt-6-astra) | Provider Adapter · high reasoning | OpenAI | 99.95% |
| 2 | [GPT-6 Astra - Provider Adapter (Max)](/models/gpt-6-astra) | Provider Adapter · max reasoning | OpenAI | 98.55% |
| 3 | [GPT-6 Astra - Provider Adapter (XHigh)](/models/gpt-6-astra) | Provider Adapter · xhigh reasoning | OpenAI | 98.44% |
| 4 | [GPT-6 Astra - Provider Adapter (Medium)](/models/gpt-6-astra) | Provider Adapter · medium reasoning | OpenAI | 98.44% |
| 5 | [GPT-6 Astra - Provider Adapter (Low)](/models/gpt-6-astra) | Provider Adapter · low reasoning | OpenAI | 98.03% |
| 6 | [Gemini 4 Argon - Provider Adapter (Low)](/models/gemini-4-argon) | Provider Adapter · low reasoning | Google | 97.46% |
| 7 | [Gemini 4 Argon - Provider Adapter (Medium)](/models/gemini-4-argon) | Provider Adapter · medium reasoning | Google | 97.34% |
| 8 | [Gemini 4 Argon - Provider Adapter (High)](/models/gemini-4-argon) | Provider Adapter · high reasoning | Google | 97.30% |
| 9 | [GPT-6 Astra - Provider Adapter (None)](/models/gpt-6-astra) | Provider Adapter · none reasoning | OpenAI | 96.72% |
| 10 | [GPT-6.1 Sol - Provider Adapter (XHigh)](/models/gpt-6-1-sol) | Provider Adapter · xhigh reasoning | OpenAI | 96.37% |
| 11 | [GPT-6.1 Sol - Provider Adapter (Max)](/models/gpt-6-1-sol) | Provider Adapter · max reasoning | OpenAI | 96.18% |
| 12 | [GPT-6.1 Sol - Provider Adapter (High)](/models/gpt-6-1-sol) | Provider Adapter · high reasoning | OpenAI | 94.99% |
| 13 | [GPT-6.1 Sol - Provider Adapter (Medium)](/models/gpt-6-1-sol) | Provider Adapter · medium reasoning | OpenAI | 91.02% |
| 14 | [GPT-6.1 Sol - Provider Adapter (Low)](/models/gpt-6-1-sol) | Provider Adapter · low reasoning | OpenAI | 82.79% |
| 15 | [Claude Sonnet 5.5 - Provider Adapter (XHigh)](https://arcprize.org/results/anthropic-claude-sonnet-5-5) | Provider Adapter · xhigh reasoning | Anthropic | 72.44% |
| 16 | [Claude Sonnet 5.5 - Provider Adapter (High)](https://arcprize.org/results/anthropic-claude-sonnet-5-5) | Provider Adapter · high reasoning | Anthropic | 35.48% |
| 17 | [Gemini 3.8 Flash - Provider Adapter (High)](/models/gemini-3-8-flash) | Provider Adapter · high reasoning | Google | 35.00% |
| 18 | [Gemini 3.8 Flash - Provider Adapter (Medium)](/models/gemini-3-8-flash) | Provider Adapter · medium reasoning | Google | 24.20% |
| 19 | [GPT-6 Sol - Provider Adapter (Max)](/models/gpt-6-sol) | Provider Adapter · max reasoning | OpenAI | 23.03% |
| 20 | [Gemini 3.8 Flash - Provider Adapter (Low)](/models/gemini-3-8-flash) | Provider Adapter · low reasoning | Google | 15.07% |
| 21 | [Grok 4.7 - Provider Adapter (High)](/models/grok-4-7) | Provider Adapter · high reasoning | xAI | 10.05% |
| 22 | [GPT-6 Sol - Provider Adapter (XHigh)](/models/gpt-6-sol) | Provider Adapter · xhigh reasoning | OpenAI | 9.45% |
| 23 | [Grok 4.7 - Provider Adapter (Medium)](/models/grok-4-7) | Provider Adapter · medium reasoning | xAI | 8.37% |
| 24 | [Grok 4.7 - Provider Adapter (XHigh)](/models/grok-4-7) | Provider Adapter · xhigh reasoning | xAI | 7.02% |
| 25 | [GPT-6 Sol - Provider Adapter (High)](/models/gpt-6-sol) | Provider Adapter · high reasoning | OpenAI | 5.86% |
| 26 | [GPT-6 Sol - Provider Adapter (Medium)](/models/gpt-6-sol) | Provider Adapter · medium reasoning | OpenAI | 4.93% |
| 27 | [Claude Sonnet 5.5 - Provider Adapter (Medium)](https://arcprize.org/results/anthropic-claude-sonnet-5-5) | Provider Adapter · medium reasoning | Anthropic | 4.89% |
| 28 | [DeepSeek V4.1 Flash - Provider Adapter (Low)](https://arcprize.org/results/deepseek-v4-1-flash) | Provider Adapter · low reasoning | DeepSeek | 3.26% |
| 29 | [DeepSeek V4.1 Flash - Provider Adapter (High)](https://arcprize.org/results/deepseek-v4-1-flash) | Provider Adapter · high reasoning | DeepSeek | 3.15% |
| 30 | [Claude Haiku 5.5 - Provider Adapter (Max)](/models/claude-haiku-5-5) | Provider Adapter · max reasoning | Anthropic | 2.82% |
| 31 | [DeepSeek V4.1 Flash - Provider Adapter (Max)](https://arcprize.org/results/deepseek-v4-1-flash) | Provider Adapter · max reasoning | DeepSeek | 2.68% |
| 32 | [GPT-6 Sol - Provider Adapter (Low)](/models/gpt-6-sol) | Provider Adapter · low reasoning | OpenAI | 1.76% |
| 33 | [GPT-6 Sol - Provider Adapter (None)](/models/gpt-6-sol) | Provider Adapter · none reasoning | OpenAI | 1.21% |
| 34 | [Grok 4.7 - Provider Adapter (Low)](/models/grok-4-7) | Provider Adapter · low reasoning | xAI | 0.77% |
| 35 | [Claude Sonnet 5.5 - Provider Adapter (Low)](https://arcprize.org/results/anthropic-claude-sonnet-5-5) | Provider Adapter · low reasoning | Anthropic | 0.62% |
| 36 | [GPT-6 Luna - Provider Adapter (Max)](/models/gpt-6-luna) | Provider Adapter · max reasoning | OpenAI | 0.59% |
| 37 | [GPT-6 Luna - Provider Adapter (XHigh)](/models/gpt-6-luna) | Provider Adapter · xhigh reasoning | OpenAI | 0.51% |
| 38 | [DeepSeek V4.1 Flash - Provider Adapter (None)](https://arcprize.org/results/deepseek-v4-1-flash) | Provider Adapter · none reasoning | DeepSeek | 0.48% |
| 39 | [Muse Spark Open - Provider Adapter (XHigh)](https://arcprize.org/results/meta-muse-spark-open) | Provider Adapter · xhigh reasoning | Meta AI | 0.42% |
| 40 | [GPT-6 Luna - Provider Adapter (High)](/models/gpt-6-luna) | Provider Adapter · high reasoning | OpenAI | 0.38% |
| 41 | [GPT-6 Luna - Provider Adapter (Medium)](/models/gpt-6-luna) | Provider Adapter · medium reasoning | OpenAI | 0.32% |
| 42 | [GPT-6 Luna - Provider Adapter (Low)](/models/gpt-6-luna) | Provider Adapter · low reasoning | OpenAI | 0.24% |
| 43 | [GPT-6 Luna - Provider Adapter (None)](/models/gpt-6-luna) | Provider Adapter · none reasoning | OpenAI | 0.05% |

## FAQ

### What does ARC-AGI-3 Provider Adapter measure?

ARC Prize verified interactive reasoning results using a provider-specific stateful adapter, kept separate from Standard harness results.

### Which model leads the published ARC-AGI-3 Provider Adapter snapshot?

GPT-6 Astra - Provider Adapter (High) currently leads the published ARC-AGI-3 Provider Adapter snapshot with a score of 99.95%.

### How many models are evaluated on ARC-AGI-3 Provider Adapter?

The Captured 2026-10-28 contains 43 AI models.

### Does ARC-AGI-3 Provider Adapter affect BenchLM's overall score?

Not directly. ARC-AGI-3 Provider Adapter is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.