ARC-AGI-3 Semi-Private: Provider Adapter harness (ARC-AGI-3 Provider Adapter)
We show this table for reference; we do not rank on it.
ARC Prize verified interactive reasoning results using a provider-specific stateful adapter, kept separate from Standard harness results.
Semi-Private score on ARC-AGI-3 Provider Adapter — Captured 2026-10-28
We mirror the published semi-private score view for ARC-AGI-3 Provider Adapter. GPT-6 Astra - Provider Adapter (High) leads the public snapshot at 99.95%, followed by GPT-6 Astra - Provider Adapter (Max) (98.55%) and GPT-6 Astra - Provider Adapter (XHigh) (98.44%). We do not use these results to rank models overall.
GPT-6 Astra - Provider Adapter (High)
OpenAI
Provider Adapter · high reasoning
GPT-6 Astra - Provider Adapter (Max)
OpenAI
Provider Adapter · max reasoning
GPT-6 Astra - Provider Adapter (XHigh)
OpenAI
Provider Adapter · xhigh reasoning
43 configurationsReasoningCurrentDisplay onlyUpdated Captured 2026-10-28
Semi-Private score table (43 configurations)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | GPT-6 Astra - Provider Adapter (High)OpenAIProvider Adapter · high reasoning | 99.95% | Not reported | Closed |
| 2 | GPT-6 Astra - Provider Adapter (Max)OpenAIProvider Adapter · max reasoning | 98.55% | Not reported | Closed |
| 3 | GPT-6 Astra - Provider Adapter (XHigh)OpenAIProvider Adapter · xhigh reasoning | 98.44% | Not reported | Closed |
| 4 | GPT-6 Astra - Provider Adapter (Medium)OpenAIProvider Adapter · medium reasoning | 98.44% | Not reported | Closed |
| 5 | GPT-6 Astra - Provider Adapter (Low)OpenAIProvider Adapter · low reasoning | 98.03% | Not reported | Closed |
| 6 | Gemini 4 Argon - Provider Adapter (Low)GoogleProvider Adapter · low reasoning | 97.46% | Not reported | Closed |
| 7 | Gemini 4 Argon - Provider Adapter (Medium)GoogleProvider Adapter · medium reasoning | 97.34% | Not reported | Closed |
| 8 | Gemini 4 Argon - Provider Adapter (High)GoogleProvider Adapter · high reasoning | 97.30% | Not reported | Closed |
| 9 | GPT-6 Astra - Provider Adapter (None)OpenAIProvider Adapter · none reasoning | 96.72% | Not reported | Closed |
| 10 | GPT-6.1 Sol - Provider Adapter (XHigh)OpenAIProvider Adapter · xhigh reasoning | 96.37% | Not reported | Closed |
| 11 | GPT-6.1 Sol - Provider Adapter (Max)OpenAIProvider Adapter · max reasoning | 96.18% | Not reported | Closed |
| 12 | GPT-6.1 Sol - Provider Adapter (High)OpenAIProvider Adapter · high reasoning | 94.99% | Not reported | Closed |
| 13 | GPT-6.1 Sol - Provider Adapter (Medium)OpenAIProvider Adapter · medium reasoning | 91.02% | Not reported | Closed |
| 14 | GPT-6.1 Sol - Provider Adapter (Low)OpenAIProvider Adapter · low reasoning | 82.79% | Not reported | Closed |
| 15 | Claude Sonnet 5.5 - Provider Adapter (XHigh)AnthropicProvider Adapter · xhigh reasoning | 72.44% | Not reported | Unknown |
| 16 | Claude Sonnet 5.5 - Provider Adapter (High)AnthropicProvider Adapter · high reasoning | 35.48% | Not reported | Unknown |
| 17 | Gemini 3.8 Flash - Provider Adapter (High)GoogleProvider Adapter · high reasoning | 35.00% | Not reported | Closed |
| 18 | Gemini 3.8 Flash - Provider Adapter (Medium)GoogleProvider Adapter · medium reasoning | 24.20% | Not reported | Closed |
| 19 | GPT-6 Sol - Provider Adapter (Max)OpenAIProvider Adapter · max reasoning | 23.03% | Not reported | Closed |
| 20 | Gemini 3.8 Flash - Provider Adapter (Low)GoogleProvider Adapter · low reasoning | 15.07% | Not reported | Closed |
| 21 | Grok 4.7 - Provider Adapter (High)xAIProvider Adapter · high reasoning | 10.05% | Not reported | Closed |
| 22 | GPT-6 Sol - Provider Adapter (XHigh)OpenAIProvider Adapter · xhigh reasoning | 9.45% | Not reported | Closed |
| 23 | Grok 4.7 - Provider Adapter (Medium)xAIProvider Adapter · medium reasoning | 8.37% | Not reported | Closed |
| 24 | Grok 4.7 - Provider Adapter (XHigh)xAIProvider Adapter · xhigh reasoning | 7.02% | Not reported | Closed |
| 25 | GPT-6 Sol - Provider Adapter (High)OpenAIProvider Adapter · high reasoning | 5.86% | Not reported | Closed |
| 26 | GPT-6 Sol - Provider Adapter (Medium)OpenAIProvider Adapter · medium reasoning | 4.93% | Not reported | Closed |
| 27 | Claude Sonnet 5.5 - Provider Adapter (Medium)AnthropicProvider Adapter · medium reasoning | 4.89% | Not reported | Unknown |
| 28 | DeepSeek V4.1 Flash - Provider Adapter (Low)DeepSeekProvider Adapter · low reasoning | 3.26% | Not reported | Unknown |
| 29 | DeepSeek V4.1 Flash - Provider Adapter (High)DeepSeekProvider Adapter · high reasoning | 3.15% | Not reported | Unknown |
| 30 | Claude Haiku 5.5 - Provider Adapter (Max)AnthropicProvider Adapter · max reasoning | 2.82% | Not reported | Closed |
| 31 | DeepSeek V4.1 Flash - Provider Adapter (Max)DeepSeekProvider Adapter · max reasoning | 2.68% | Not reported | Unknown |
| 32 | GPT-6 Sol - Provider Adapter (Low)OpenAIProvider Adapter · low reasoning | 1.76% | Not reported | Closed |
| 33 | GPT-6 Sol - Provider Adapter (None)OpenAIProvider Adapter · none reasoning | 1.21% | Not reported | Closed |
| 34 | Grok 4.7 - Provider Adapter (Low)xAIProvider Adapter · low reasoning | 0.77% | Not reported | Closed |
| 35 | Claude Sonnet 5.5 - Provider Adapter (Low)AnthropicProvider Adapter · low reasoning | 0.62% | Not reported | Unknown |
| 36 | GPT-6 Luna - Provider Adapter (Max)OpenAIProvider Adapter · max reasoning | 0.59% | Not reported | Closed |
| 37 | GPT-6 Luna - Provider Adapter (XHigh)OpenAIProvider Adapter · xhigh reasoning | 0.51% | Not reported | Closed |
| 38 | DeepSeek V4.1 Flash - Provider Adapter (None)DeepSeekProvider Adapter · none reasoning | 0.48% | Not reported | Unknown |
| 39 | Muse Spark Open - Provider Adapter (XHigh)Meta AIProvider Adapter · xhigh reasoning | 0.42% | Not reported | Unknown |
| 40 | GPT-6 Luna - Provider Adapter (High)OpenAIProvider Adapter · high reasoning | 0.38% | Not reported | Closed |
| 41 | GPT-6 Luna - Provider Adapter (Medium)OpenAIProvider Adapter · medium reasoning | 0.32% | Not reported | Closed |
| 42 | GPT-6 Luna - Provider Adapter (Low)OpenAIProvider Adapter · low reasoning | 0.24% | Not reported | Closed |
| 43 | GPT-6 Luna - Provider Adapter (None)OpenAIProvider Adapter · none reasoning | 0.05% | Not reported | Closed |
How to read this leaderboard
Operator receipt: 43 sourced rows are currently displayable on this page; the leading published row is GPT-6 Astra - Provider Adapter (High) at 99.95%.
Honest limit: A model-plus-adapter result is a complete-system configuration. Do not pool it with Standard harness results, change evaluation splits, or infer unpublished runs. Display-only and excluded from weighted ranking.
Provider Adapter configurations
These are ARC Prize's published Semi-Private results. Every published reasoning-effort configuration stays separate; the table does not combine efforts or evaluation splits.
Standard and Provider Adapter harnesses are separate protocols. The Provider Adapter retains opaque reasoning state and compacts long conversations. A score gap can reflect those system differences.
Model pages use the highest declared published reasoning effort within each harness, rather than selecting whichever effort scored highest. Hidden or unpublished incomplete runs are not assigned a score.
The published ARC-AGI-3 Provider Adapter snapshot places GPT-6 Astra - Provider Adapter (High) first at 99.95%. The third row is 1.50 points behind. The broader top-10 range is 3.57 points, so many of the published results sit in a relatively narrow band.
43 configurations have been evaluated on ARC-AGI-3 Provider Adapter. The benchmark falls in the Reasoning category. ARC-AGI-3 Provider Adapter is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About ARC-AGI-3 Provider Adapter
Year
2026
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.
Freshness and provenance
Version
ARC-AGI-3 Provider Adapter 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
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 99.95% semi-private score. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on ARC-AGI-3 Provider Adapter?
The Captured 2026-10-28 snapshot contains 43 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.