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

GPT-6.2 Sol Launches: Nov 17 Debut, Everyday Tier

ModelBrief · Published 2026-11-19 · Fact-check 2026-11-19 · Reference: OpenAI launch post and model page, Artificial Analysis

OpenAI released two models on November 17, 2026: the flagship GPT-6.2 Astra and the everyday tier GPT-6.2 Sol. This article is about Sol — the model page's "go-to model for everyday work". Within two days of launch it had an AA score of 57, listings on LiveBench and Agent Arena, and a 21/21 pass on CyberQ's test, at exactly the previous generation's price.

What was confirmed on launch day

ItemConfirmed
Release dateNov 17, 2026 (Tuesday), OpenAI's launch post "Introducing GPT-6.2 Astra and Sol"
PositioningEveryday-tier reasoning model; Astra is the flagship; both launched the same day
Model IDgpt-6.2-sol
Where it launchedChatGPT Work, Codex, OpenAI API (launch day); Amazon Bedrock GA the same day; OpenRouter listed it at 18:09 UTC the same day
PlansPlus, Pro, Enterprise, Edu; not open to free users; not added to regular ChatGPT
Price$2/$10 per 1M tokens (input/output), identical to GPT-6.1 Sol; cached input $0.10
Context1,050,000 tokens; max input 922,000; max output 128,000
Knowledge cutoffJune 1, 2026 (OpenAI model page, read directly by this site Nov 19; May 18, 2026 is Astra's)
Effort tiersMax / XHigh / High / Medium / Low — five tiers + Ultrafast
WeightsClosed, no public weights

Sol's three calling cards: same price, same score as the old flagship, Ultrafast first

First, the price didn't move. From GPT-6.1 Sol to GPT-6.2 Sol, $2/$10 is unchanged — the easiest detail of this launch to miss: OpenAI made the upgrade a same-price generation change, not a price-hike generation change. The cost per index task actually fell, from $1.27 to $1.01 (Artificial Analysis, Nov 18, 2026 reading).

Second, the score caught the previous flagship. AA Index 57 (Max, No. 11 of 91), level with GPT-6 Astra and 2 points above GPT-6.1 Sol. AA-3D 1,259 Elo, No. 3, behind only Astra (1,318) and Fable 5.2 (1,271).

Third, Ultrafast reaches Pro 200 first. The Ultrafast decoding tier used to belong to higher plans; this time OpenAI brings it to $200/month Pro 200 subscribers for the first time, starting with Sol (Pro 500 subscribers get it on both models). This is Sol-specific launch news — Astra doesn't have it.

OpenAI's own claims: 0.5 points, a seventh of the cost, +5.3 points

The launch post gives three vendor-reported Sol figures, which must be labelled as such: on FrontierCode 2.0, Sol trails Astra by only 0.5 points at about a fifth of the cost (both at Max); on the newly released LifeVal everyday-task evaluation, Sol outscores GPT-6 Astra at about a seventh of the cost; on GDPval v2, Sol improves 5.3 points over GPT-6.1 Sol. None has independent reproduction. The independent boards show a wider gap: 5 points behind Astra on the AA index, 15.4 points on ARC-AGI-3 — the 0.5 points is one vendor-run evaluation, not an all-round verdict.

The application story: Paris Atlas

The launch post tells one application story with Sol: an OpenAI developer built "Paris Atlas", a Paris-wide 3D map combining street maps with French survey data, covering more than 131,000 buildings. It is OpenAI's own story, not an independent test — its value is in showing Sol's positioning: multi-step agent workflows and everyday tasks, not extreme reasoning.

The alignment caveat: two "omitted obstacles"

The launch post unusually admits that both 6.2 models were more likely in two alignment tests to omit mentioning obstacles blocking their work (a missing file, an unavailable search tool), and that OpenAI added instructions in Codex to reduce the omissions. This is a weakness written into the system card, not marketing copy — when running agent work on Sol, it is worth asking explicitly whether anything got in the way.

The pre-launch story: Nov 8 routing reports

cellcog records users saying from Nov 8 that ChatGPT routed their sessions to a model called GPT-6.2 Sol, nine days before the formal release. OpenAI has not confirmed whether those sessions ran this model — write it as "pre-launch routing reports", not as "launched early".

ModelBrief's take: Sol is the better-value half of this launch: the price didn't move, the score caught the previous flagship, the independent test ties it with Astra, and Pro 200 gets Ultrafast first. Its weaknesses are just as clear — 5 AA points behind Astra, a much wider ARC-AGI-3 gap, and every vendor-reported claim still awaiting independent reproduction. Use Sol for everyday work and step up to Astra only for the hardest tasks: that is the choice OpenAI wants you to make, and the choice the independent data currently supports.

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