AI Trends

Trending story

ARC Prize announces Mercury 3 Preview's evaluation results on ARC-AGI

3 reports3 story updatesUpdated Nov 17 23:02

Understand this story

AI overview

On November 17, 2026, ARC Prize announced the evaluation results of Inception's Mercury 3 Preview on ARC-AGI. It scored 88.5% on ARC-AGI-1 at $0.03 per task; 61.8% on ARC-AGI-2 at $0.06 per task; and 0.2% on ARC-AGI-3. ARC Prize stated that this is the first diffusion model it has evaluated, with an average output of 981 tokens/second in instant reasoning mode and a peak of 1,743 tokens/second for a single call; with reasoning enabled, the average is 500-600 tokens/second. Another ARC Prize publication also lists ARC-AGI-3 at 0.2% ($299) under the provider adapter and 0.1% ($333) under standard. There are currently no further developments.

AI-generated from public reports

Latest development · Report

ARC Prize announces evaluation results of Mercury 3 Preview on ARC-AGI

Developments

1 confirmed developments

  1. ARC Prize announces evaluation results of Mercury 3 Preview on ARC-AGI

    Source:ARC Prize

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Nov 17

  1. ARC PrizeFirst-party

    ARC Prize announces evaluation results of Mercury 3 Preview on ARC-AGI

    ARC Prize announced the evaluation results of Inception's Mercury 3 Preview on ARC-AGI: a score of 88.5% on ARC-AGI-1 at $0.03 per task, 61.8% on ARC-AGI-2 at $0.06 per task, and 0.2% on ARC-AGI-3. This is the first diffusion model evaluated by ARC Prize, averaging 981 tokens/second output in instant reasoning mode with a peak of 1,743 tokens/second for a single call; with reasoning enabled, it averages 500-600 tokens/second.

  2. ARC PrizeFirst-party

    ARC Prize's measured results for Mercury 3 Preview reasoning tiers and token consumption

    On the ARC-AGI-1 and ARC-AGI-2 tasks, ARC Prize observed that the low and medium reasoning tiers of Mercury 3 Preview consume similar numbers of reasoning tokens. On the ARC-AGI-1 public tasks, the two tiers each used an average of 31k reasoning tokens per task, with scores of 89.50% and 89.25%, respectively. ARC Prize previously reported that Mercury 3 Preview scored 61.8% on ARC-AGI-2 at a cost of $0.06 per task, and scored 88.5% on ARC-AGI-1 at a cost of $0.03 per task.

  3. ARC PrizeFirst-party

    ARC Prize announces that Inception AI's Mercury 3 Preview on…

    ARC Prize announces the validation results of Inception AI's Mercury 3 Preview on ARC-AGI: ARC-AGI-1 score 88.5%, $0.03 per task; ARC-AGI-2 score 61.8%, $0.06 per task; ARC-AGI-3 score 0.2% (provider adapter, $299) and 0.1% (standard, $333). This is the first diffusion model evaluated by ARC Prize, with a maximum output speed of 1,743 tokens/second; at the instant reasoning tier, the ARC-AGI-2 public set averages 981 output tokens/second, with a single-call peak of 1,743 tokens/second (3,001 tokens output in 1.72 seconds), and averages 500-600 tokens/second with reasoning enabled.

Related stories

  1. Related storiesGoogle launches Gemini 4 Hydrogen on Antigravity
  2. Related storiesThinking Machines Lab releases Inkling 2 Small
  3. Related storiesGoogle DeepMind releases the Gemini 4 Carbon model
  4. Related storiesMeta launches Muse Wonder and Muse Wonder Full models