RLTasksby ulam.ai
Models / OpenAI

OpenAI / MODEL EVIDENCE

GPT-6 Astra (none)

5 benchmarks · 5 observations · 5 sources · released 2026-09-03

Scores at a glance

5 benchmarks · large dot: GPT-6 Astra (none); small dots: closest models where reported

CritPt ↗19.71%
032.29

GPT-6 Astra (none) · CritPt · Accuracy · % ↗

SciCode ↗53.47%
1.2766.9

GPT-6 Astra (none) · SciCode · Score · % ↗

Closest models by score

Similar positions on shared benchmark leaderboards. Smaller gaps are closer.

3.6 percentile-point gap

5 of 5 shared benchmark comparisons

How similarity is measured

Mean absolute gap in percentile positions on shared benchmark settings; lower is closer. At least 60% overlap is required, with a minimum of three when available. Independent evaluations are used when there are at least three; otherwise published report comparisons are used. Repeated model observations within one setting are averaged before ranking. Writing-style traits and composite indices are excluded. This describes similarity, not an overall quality ranking.

POST-TRAINING / SOURCE DISCLOSURE

What the GPT-6 Astra report says

The system card discusses reinforcement-learning grading, alignment improvements and safety behavior integrated into post-training. It reports how these changes are evaluated rather than providing the full production training recipe.

  • Reinforcement learning with alignment-focused grading
  • Safety behavior integrated into post-training
Preference learning and RLHF · background →Reward design · background →

A specific optimizer, training-data mixture and complete compute ledger are not given in the reviewed system card. Evaluation scores do not identify those missing details.

  1. GPT-6 Astra System Card ↗

    OpenAI · 2026

Filter comparisons to this model →

Sources

Every plotted observation links to its evaluation. Reported configurations, benchmark versions and task subsets can differ; similarity describes shared scores, not training lift.

  1. ARC Prize - Leaderboard ↗

    ARC Prize Foundation

  2. SciCode Benchmark | Artificial Analysis ↗

    Artificial Analysis

  3. ARC-AGI-2 ↗

    Epoch AI

  4. CritPt v1 ↗

    Epoch AI

  5. FrontierMath Tier 4 (v2) ↗

    Epoch AI

Names used in source tables

GPT-6 Astra (none)