Checked Nov 10, 2026Start a new check

What was checked

Claude Fable 5.2 from Anthropic sets a new high score on ARC-AGI-3: 97.7% (standard harness)

Assessment summary

Inaccurate

Claude Fable 5.2 did achieve 97.7% on ARC-AGI-3, but this score was with the Provider Adapter harness, not the standard harness. The claim omits this critical distinction. With the standard harness, Fable 5.2 scored 81.2%, which is the actual new high score for the standard harness.

Key findings

Model assessments

Assessment summary

  • The claim is contradicted by available benchmark reporting.
  • The reported 97.72% score was on a Provider Adapter harness, while the Standard harness score was reported as 65.66%, not 97.7%.llm-stats.com reconscribe.com

Claims reviewed

Claude Fable 5.2 from Anthropic sets a new high score on ARC-AGI-3: 97.7% (standard harness)

Contradicted
The 97.72% result is reported for the Provider Adapter harness, not the Standard harness.llm-stats.com reconscribe.com The Standard harness result is reported as 65.66%, so the claim mislabels the score and its evaluation setup.llm-stats.com

This claim may mislead readers without added context.

Assessment summary

  • The claim contains a factual inaccuracy regarding which harness achieved the 97.7% score.
  • Claude Fable 5.2 did achieve 97.7% on ARC-AGI-3, but this was with the Provider Adapter harness, not the standard harness.
  • With the standard harness, Fable 5.2 scored 81.2%, which is indeed a new high score for that harness.
  • The 97.7% score is misleading without the harness specification.

Claims reviewed

Claude Fable 5.2 from Anthropic sets a new high score on ARC-AGI-3: 97.7%

Needs context
Claude Fable 5.2 did achieve 97.7% on ARC-AGI-3, but this score was with the Provider Adapter harness, not the standard harness. The claim omits this critical distinction. With the standard harness, Fable 5.2 scored 81.2%, which is the actual new high score for the standard harness.

This claim may mislead readers without added context.

Claude Fable 5.2 sets a new high score on ARC-AGI-3 in the standard harness

Supported
Claude Fable 5.2 does set a new high score on ARC-AGI-3 with the standard harness at 81.2%, surpassing GPT-6 Astra's 62.7%. This is verified by ARC Prize and multiple sources.

Assessment summary

  • The statement mixes up the harness for the 97.7% number.
  • ARC Prize's verified results show Claude Fable 5.2 reached 97.7% on the Provider Adapter harness, not the Standard harness arcprize.org.
  • The Standard-harness scores for Fable 5.2 are much lower (e.g., 81.2% at XHigh, 79.3% at High) so the claim as written is incorrect arcprize.org handyai.substack.com.

Claims reviewed

97.7% (standard harness)

Contradicted
ARC Prize's verified page shows 97.7% under the Provider Adapter harness (High reasoning); the Standard harness values are 81.2% (XHigh) and 79.3% (High), not 97.7% arcprize.org.

This claim may mislead readers without added context.

Claude Fable 5.2 from Anthropic sets a new high score on ARC-AGI-3: 97.7%

Needs context
Fable 5.2 does have a 97.7% result on ARC-AGI-3, but that score is reported for the Provider Adapter harness, not the Standard harness arcprize.org handyai.substack.com.

This claim may mislead readers without added context.

Sources cited

14 sources

Share this result