Tech Meridian← ALL MODELS
PROMY MERIDIAN

OPENAI · MODEL RELEASE TRACKER

GPT-6.2 Sol

GPT-6.2 Sol is a version of OpenAI's GPT-6.2 family. An evaluation by Artificial Analysis reported that Sol produces substantially shorter, more direct answers with much less structural formatting than its predecessor, and it admits uncertainty far less often.

CURRENT SNAPSHOT4/5 DIMENSIONS WITH DATA

The dimensions that change the decision.

PRICING
API pricingReported to keep GPT‑6.1 Sol's pricing: $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens.
CONTEXT WINDOW

Not established from the available sources.

MODALITIES
ModalityDescribed as a new OpenAI language model (part of the GPT‑6.2 release).
BENCHMARKS
Average answer length (words)Dropped from 337 to 223 words (34% decrease) compared to its predecessor, per Artificial Analysis.
Average number of headings per answerFell from 1.85 to 0.25 headings per answer.
Organization IndexDropped from 86 to 27 on the Organization Index (a composite of structure, clarity, and related traits).
Answer First trait scoreIncreased from 64 to 87 (indicating the model more often addresses the actual question immediately).
Natural sounding language scoreRose from 54 to 79.
Comparison vs Anthropic (structure/transparency)Rated lower on structure and transparency compared with Anthropic's Claude Fable 5.2 and Opus 5.5 in the reported evaluation.
AVAILABILITY
Availability / accessReported as available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex; accessible via the OpenAI API. Pro‑500 subscribers can use Ultrafast mode with both GPT‑6.2 models, and Pro‑200 subscribers receive Ultrafast access starting with GPT‑6.2 Sol.

RELEASE TIMELINE

Published, source-backed release events only.

2 SOURCES · IMPORTANCE 9.0

OpenAI introduces GPT-6.2 Astra and Sol

OpenAI announced GPT‑6.2, released as two variants — Astra and Sol — promising improved professional outputs, coding, and computer use with greater efficiency and lower cost across ChatGPT Work, Codex, and the API. The update is positioned to deliver better performance for enterprise and developer workflows.

→

VERIFIABLE FACTS

Every value stays attached to a source and date.

AVAILABILITY · Availability / accessDEVELOPER CLAIM

Reported as available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex; accessible via the OpenAI API. Pro‑500 subscribers can use Ultrafast mode with both GPT‑6.2 models, and Pro‑200 subscribers receive Ultrafast access starting with GPT‑6.2 Sol.

LIMITATIONS · API behavior: editing history discards earlier reasoning; some parameters removedDEVELOPER CLAIM

OpenAI notes that in the API for GPT-6.2 (Astra and Sol), editing conversation history can discard earlier reasoning steps — which may affect token costs and response quality — and parameters such as 'temperature' and 'top_p' are removed.

WHAT CHANGED

Stored passport versions, without reconstructed history.

Passport updated18 facts
SAFETY · Admits Uncertainty trait score+ Fell from 98 to 51, indicating the model admits uncertainty far less often in the evaluation.
BENCHMARKS · Answer First trait score+ Increased from 64 to 87 (indicating the model more often addresses the actual question immediately).
BENCHMARKS · Average answer length (words)+ Dropped from 337 to 223 words (34% decrease) compared to its predecessor, per Artificial Analysis.
BENCHMARKS · Comparison vs Anthropic (structure/transparency)+ Rated lower on structure and transparency compared with Anthropic's Claude Fable 5.2 and Opus 5.5 in the reported evaluation.
SAFETY · Tightened safeguards against extracting internal reasoning− OpenAI has tightened safeguards to make it much harder for attackers to extract internal reasoning from GPT-6.2 models, including Sol; defenses include targeted training, input/output filters, API blocks that limit older models' access to protected reasoning, and account bans to disrupt coordinated attacks.
BENCHMARKS · Average number of headings per answer+ Fell from 1.85 to 0.25 headings per answer.
BENCHMARKS · Natural sounding language score+ Rose from 54 to 79.
BENCHMARKS · Organization Index+ Dropped from 86 to 27 on the Organization Index (a composite of structure, clarity, and related traits).
Passport updated12 facts
LIMITATIONS · API behavior: editing history discards earlier reasoning; some parameters removed+ OpenAI notes that in the API for GPT-6.2 (Astra and Sol), editing conversation history can discard earlier reasoning steps — which may affect token costs and response quality — and parameters such as 'temperature' and 'top_p' are removed.
LIMITATIONS · Tendency to answer briefly and make assumptions+ OpenAI reports that GPT-6.2 models, including Sol, answer more briefly by default and should be prompted to make reasonable assumptions rather than interrupt workflows with clarifying questions; developers are advised to adjust system prompts accordingly.
SAFETY · Tightened safeguards against extracting internal reasoning+ OpenAI has tightened safeguards to make it much harder for attackers to extract internal reasoning from GPT-6.2 models, including Sol; defenses include targeted training, input/output filters, API blocks that limit older models' access to protected reasoning, and account bans to disrupt coordinated attacks.
RELEASE · New prompt templates released+ OpenAI released new prompt templates and copy-ready prompts in the documentation for GPT-6.2 Astra and GPT-6.2 Sol covering writing style, initiative, and coding, and advised developers to update system prompts when moving to these models.
CAPABILITIES · Design and cost positioning+ OpenAI describes Sol as intended to handle complex work at lower cost (contrasted with Astra, which is built for the most demanding tasks).
SAFETY · Tightened safeguards against extracting internal reasoning− OpenAI has sharply tightened safeguards to make it much harder to extract internal reasoning from GPT-6.2 Sol.
Passport updated8 facts
SAFETY · Mitigation of coordinated campaigns+ Against coordinated campaigns, OpenAI bans thousands to tens of thousands of accounts weekly as part of its defenses.
SAFETY · Individual-request defenses+ For individual requests, defenses combine targeted training, input and output filters for messages and tool calls, and API blocks that stop older models from reading protected reasoning.
SAFETY · Tightened safeguards against extracting internal reasoning+ OpenAI has sharply tightened safeguards to make it much harder to extract internal reasoning from GPT-6.2 Sol.
Passport created5 facts