DoubleChecks

What was claimed

Inkling achieves 79.5% on ARC-AGI-1 and 36.5% on ARC-AGI-2 as highest-scoring open-weight model

Our verdict

Needs Caution

Current public leaderboards and reports list other models and scores (e.g., GPT‑5 around mid‑60s on ARC‑AGI‑1, o3 up to the high‑80s, various PoE/ensemble methods in the 70s), but do not mention a model called Inkling or a 79.5% open‑weight result. Without any authoritative source referencing Inkling or this score, the claim cannot be confirmed or directly refuted. No published results show Inkling evaluated on ARC-AGI-2. The highest documented open-weight score is 24% (ARC Prize 2025 Kaggle winner). The claim's score is implausibly high.

0 of 3 AI systems agree13 sources citedChecked Jul 18, 2026

Check your own claim

Paste any statement, headline, or AI answer — 3 independent AIs verify it in seconds, with sources.

Key findings

Inkling achieves 79.5% on ARC-AGI-1 as the highest-scoring open-weight model.

Can’t verify69%
3 of 6 AIs agree·ChatGPT: Verified, ChatGPT: Verified, Perplexity: Incorrect

Inkling achieves 36.5% on ARC-AGI-2

Can’t verify85%
1 AI checked

Inkling is the highest-scoring open-weight model on these benchmarks

Verified88%
1 of 2 AIs agree·Claude: Can’t verify

Verified against 13 sources

frontiermilestones.orgarcprize.orgarcprize.orgemergentmind.comreddit.comericcampbell.devarcprize.orggenailist.netarcprize.orgnote.comgithub.comarcprize.orgarcprize.org

Detailed Analysis

The claimed ARC-AGI-1 and ARC-AGI-2 scores for an open-weight model named Inkling are not supported by any public benchmark or leaderboard. The specific numbers given are also inconsistent with current reported ranges for both ARC-AGI-1 and ARC-AGI-2 performance.

Why this verdict

  • The claimed ARC-AGI-1 and ARC-AGI-2 scores for an open-weight model named Inkling are not supported by any public benchmark or leaderboard.
  • The specific numbers given are also inconsistent with current reported ranges for both ARC-AGI-1 and ARC-AGI-2 performance.

Claims checked

Inkling is the highest-scoring open-weight model on ARC-AGI-1 and ARC-AGI-2.

Incorrect86%
Recent leaderboards and technical reports identify multiple high-performing systems and ensembles on ARC‑AGI‑1 and ARC‑AGI‑2, including commercial models, contest winners like NVARC, and open methods such as PoE+DFS, but none list Inkling or credit it as state of the art.[4] [9] [11] [13] Given these comprehensive rankings and the absence of Inkling, the assertion that it is the highest-scoring open‑weight model is contradicted by available evidence.

This claim may mislead readers without added context.

Inkling achieves 36.5% on ARC-AGI-2.

Can’t verify60%
Published ARC‑AGI‑2 results for both commercial and competition systems cluster mostly in the low‑ to mid‑20% range at the high end, with top contest systems around 24% and other well‑documented methods below that.[9] [11] [12] [13] No reliable sources mention a model called Inkling or a 36.5% score on ARC‑AGI‑2, so this specific claim cannot be verified.

Inkling achieves 79.5% on ARC-AGI-1 as the highest-scoring open-weight model.

Can’t verify62%
Current public leaderboards and reports list other models and scores (e.g., GPT‑5 around mid‑60s on ARC‑AGI‑1, o3 up to the high‑80s, various PoE/ensemble methods in the 70s), but do not mention a model called Inkling or a 79.5% open‑weight result.[2] [4] [8] [9] [11] [12] [13] Without any authoritative source referencing Inkling or this score, the claim cannot be confirmed or directly refuted.

What other AI think

The claim cannot be verified. While Inkling is a real open-weight model released July 15, 2026, no evidence exists that it has been evaluated on ARC-AGI benchmarks. The stated scores far exceed documented open-weight model performance on these benchmarks.

Why this verdict

  • The claim cannot be verified.
  • While Inkling is a real open-weight model released July 15, 2026, no evidence exists that it has been evaluated on ARC-AGI benchmarks.
  • The stated scores far exceed documented open-weight model performance on these benchmarks.

Claims checked

Inkling achieves 79.5% on ARC-AGI-1

Can’t verify85%
No published results show Inkling evaluated on ARC-AGI-1. The highest documented open-weight score is ~45% (TRM). The claim's score is implausibly high for an open-weight model.

Inkling achieves 36.5% on ARC-AGI-2

Can’t verify85%
No published results show Inkling evaluated on ARC-AGI-2. The highest documented open-weight score is 24% (ARC Prize 2025 Kaggle winner). The claim's score is implausibly high.

Inkling is the highest-scoring open-weight model on these benchmarks

Can’t verify90%
Without verified ARC-AGI scores for Inkling, this claim cannot be assessed. Inkling was released July 15, 2026, very recently, and no ARC-AGI benchmark results are documented.
The numeric scores are reported by ARC Prize and by the model publisher. ARC Prize lists Inkling at 79.5% on ARC-AGI-1 and 36.5% on ARC-AGI-2. ARC Prize also indicates Inkling is the top-performing open-weight model on their verified leaderboard.

Why this verdict

  • The numeric scores are reported by ARC Prize and by the model publisher.
  • ARC Prize lists Inkling at 79.5% on ARC-AGI-1 and 36.5% on ARC-AGI-2.
  • ARC Prize also indicates Inkling is the top-performing open-weight model on their verified leaderboard.

Claims checked

Inkling is the highest-scoring open-weight model.

Verified88%
ARC Prize’s verified leaderboard and the ARC Prize result post state Inkling is the highest-scoring open-weight model on their evaluated ARC-AGI sets.

Inkling achieves 36.5% on ARC-AGI-2.

Verified90%
ARC Prize’s verified results and the Thinking Machines announcement list a 36.5% score for Inkling on ARC-AGI-2.

Inkling achieves 79.5% on ARC-AGI-1.

Verified92%
ARC Prize and Thinking Machines’ release report Inkling scoring 79.5% on the ARC-AGI-1 semi-private evaluation set.

Share this result

Try DoubleChecks