ModelBrief · GPT-6.2 Astra · Topic Page
Last updated Nov 20, 2026

GPT-6.2 Astra Benchmarks: AA 62 and the Rank Puzzle

ModelBrief · Published 2026-11-18 · Fact-check 2026-11-18 · Reference: Artificial Analysis, OpenAI launch post (via secondary reports)

One day after launch, GPT-6.2 Astra's benchmark picture is still thin: Artificial Analysis is the only independent yardstick so far, and OpenAI's two self-reported leads have no independent reproduction yet. This article tallies every checkable score — first what the 62 is made of, then why the rank has two wordings, then which numbers may only be cited as vendor figures.

AA 62: what it is made of, and what it costs

Artificial Analysis Intelligence Index v5.0.1 (Max tier): Astra 62, top 3 of 91 models; Claude Fable 5.2 leads at 65; GPT-6 Astra 57, GPT-6.2 Sol 57 (No. 11), GPT-6.1 Sol 55. AA measures $4.00 per index task, 11.5 minutes, 65.8M tokens generated. For comparison: Fable 5.2 costs $26.54 per task, GPT-6 Astra $6.24, Sol $1.01. Astra is one of the cheapest flagships per task near the top of the board.

GPT-6.2 Astra AA Intelligence Index scores and cost per task vs main rivals Fable 5.2 65 · $26.54/task GPT-6.2 Astra 62 · $4.00/task GPT-6 Astra 57 · $6.24/task GPT-6.2 Sol 57 · $1.01/task GPT-6.1 Sol 55 · $1.27/task Gemini 4 Carbon 60 (comparison)
AA Intelligence Index v5.0.1 (points) and cost per index task (Artificial Analysis, Nov 18, 2026 reading; grey bars are comparison models; Carbon's per-task cost was not confirmed in this round)

The rank puzzle: No. 2 or No. 3?

AA's own X post calls Astra the "#2 model in the Index", while this site's Nov 18 AI briefing says No. 3 of 91 models. The two wordings may come from different tiers, different versions or different counting bases (the board's model count changes between versions), and they cannot be reconciled yet. This page and the topic page both say "top 3"; Fable 5.2 (65) is undisputed at No. 1. This site will update once the AA model page or an official note settles the wording.

AA-3D and AnalystAgent: Astra's two firsts

AA-3D (brief-to-3D, human blind rating): Astra 1,318 Elo, No. 1, ahead of Fable 5.2's 1,271 — a direct reversal against the board leader. AA-AnalystAgent: 62.5%. Agents' Last Exam v1.1: about 78% at about $3 per task (AA X post, via aiweekly). Of the three, the AA-3D reversal matters most: Astra beats the board leader in blind-rated 3D generation.

The token-count caveat: cheaper rates, not thriftier output

metarix notes Astra generated 65.8M tokens on the AA index task, more than GPT-6 Astra's 54.3M — the per-task cost fell from $6.24 to $4.00 because rates fell ($8/$40 vs $10/$50), not because the model used fewer tokens. In other words, Astra is stronger and cheaper, but it talks more. Budget by per-task cost, not by the rate card alone.

Vendor-reported: two leads, no independent reproduction yet

OpenAI's launch post claims two leads: Sol trails Astra by only 0.5 points on Cognition FrontierCode 2.0 at about a fifth of the cost; Astra leads Fable 5.2 by 5.5 points on Agents' Last Exam at 5% of the cost (Fable about $60 per task, per aiweekly). Both are vendor figures, with rival scores re-run by OpenAI rather than each vendor's published values, and neither has an independent reproduction yet. Cite them as "vendor-reported" — never write them into a comparison table as independent measurements.

The missing boards

As of Nov 18, LMArena and LiveBench have not listed GPT-6.2 Astra; tier-1 English coverage (TechCrunch / The Verge / Reuters / Bloomberg) is unconfirmed in this round; systematic Chinese tech-media coverage is still coming. On HealthAdminBench (Stanford-developed medical-admin tasks), Astra scores 94.3% and Luna 91.8% — an application story, not a core benchmark, with sample size and method details not yet public.

ModelBrief's take: The 62 itself is undisputed; only the "No. 2 or No. 3" wording is. The real weakness is how few independent yardsticks exist — beyond AA, no second independent body has scored it yet. Astra's two firsts (AA-3D, AnalystAgent at 62.5%) and its $4.00 per-task bill are the hardest independent evidence so far; OpenAI's two self-reported leads should wait for reproduction before being believed.

Reference: Artificial Analysis — GPT-6.2 Astra model page · OpenAI — GPT-6.2 Astra model page (rewritten and organized by ModelBrief; facts per official sources and Artificial Analysis)