Day 7 · 2026-11-17
35 launches checked147 claims quoted word for word179 user reports linked10 verified fixes linked40 corrections publishedLive 25 days5 of 5 founding sponsor spots left
AI launches, checked against real usersLast re-stamped 17 Nov 2026

All receipts / No. 035

No. 035 · ModelChecked 17 Nov 2026

Day 7 receipt

GPT-6.2 Astra and Sol

OpenAI · launched 17 Nov 2026
Launch day · 0 of 7

Claims on the record

We check on 24 Nov 2026

They promised
Frontier-level results on professional work at far lower cost per task, with more done inside the same usage limits.

The claims we’ll check

  • “Both models deliver better results on demanding tasks with fewer output tokens, so you can get more done within your usage limits.”
  • “GPT‑6.2 Astra sets a new high, scoring 5.2 points higher than GPT‑6 Astra at less than a third of the cost, and 5.5 points ahead of Claude Fable 5.2 at just 5% of the cost.”
  • “GPT‑6.2 Astra sets a new state of the art, more than doubling Claude Fable 5.2’s pass rate at about one-sixth the cost.”
  • “GPT-6.2 Astra is our first externally released model to meet the High threshold for AI Self-improvement”
  • “in two tests, they were more likely to omit mentioning obstacles that blocked their work, such as a missing file or unavailable search tool.”

Quoted word for word from Introducing GPT-6.2 Astra and Sol. Day 1: the headline wins are OpenAI's own charts, and independent testing so far is mixed, with a top-two ranking overall but no gain on coding agents.

Full evidence 5 claims · 0 user reports · 3 sources

OpenAI's new model pair: Astra, a frontier model for hard professional work, and Sol, an efficient everyday model. Both are in ChatGPT Work, Codex and the API, with an Ultrafast mode reaching a second paid tier for the first time.

Launch post: Introducing GPT-6.2 Astra and Sol

Why this stamp

Day 1: the headline wins are OpenAI's own charts, and independent testing so far is mixed, with a top-two ranking overall but no gain on coding agents.

OpenAI says both models do better work on hard tasks with fewer output tokens, and that Astra leads Claude Fable 5.2 on several benchmarks at a fraction of the cost. Artificial Analysis puts Astra second on its Intelligence Index, behind Fable 5.2, and finds Astra flat and Sol 3 points lower on its coding-agent index. OfficeChai notes most of the comparisons are OpenAI's own charts and that Gemini 4 Carbon is missing from several. OpenAI also discloses that both models more often stay silent about blocked work in two tests, and that Astra is its first released model at the High self-improvement level. No user reports yet. We re-stamp on day 7 (24 November) and day 14.

Launch dayWhat they claimed, quoted word for word · 5

“Both models deliver better results on demanding tasks with fewer output tokens, so you can get more done within your usage limits.”
? UnclearEvidence: S1, S3

Artificial Analysis ranks Astra second on its own index, behind Claude Fable 5.2, and shows its coding-agent index flat for Astra and down 3 points for Sol. Usage-limit gains are untested by users.

“GPT‑6.2 Astra sets a new high, scoring 5.2 points higher than GPT‑6 Astra at less than a third of the cost, and 5.5 points ahead of Claude Fable 5.2 at just 5% of the cost.”
? UnclearEvidence: S1, S2

Agents' Last Exam result as OpenAI reports it, with nine Fable tasks excluded. OfficeChai notes Gemini 4 Carbon sits under a point behind on the same chart.

“GPT‑6.2 Astra sets a new state of the art, more than doubling Claude Fable 5.2’s pass rate at about one-sixth the cost.”
? UnclearEvidence: S2

Harvey LAB-AA is a legal benchmark. OfficeChai reads the chart at roughly 15% against 7%, both low absolute scores. No independent rerun yet.

“GPT-6.2 Astra is our first externally released model to meet the High threshold for AI Self-improvement”
? UnclearEvidence: S2, S3

OpenAI's own statement, repeated by OfficeChai and TestingCatalog. A disclosure of its own safety rating, not an independently checked one.

“in two tests, they were more likely to omit mentioning obstacles that blocked their work, such as a missing file or unavailable search tool.”
? UnclearNo linked reports

OpenAI discloses this in its own post and says it added Codex instructions that reduce it. Ongoing investigation.

OthersWhat outlets and testers say · 3

S1Artificial Analysis17 Nov 2026neutral

Independent index: Astra second behind Claude Fable 5.2, strong in science and 3D, but no gain on its coding-agent index and a 59% hallucination rate on its knowledge test.

“OpenAI’s release of GPT-6.2 Astra and GPT-6.2 Sol delivers the #2 model in the Artificial Analysis Intelligence Index” · Open the source

S2OfficeChai17 Nov 2026skeptical

Reads OpenAI's charts, notes the numbers are OpenAI's own on chosen benchmarks and that Gemini 4 Carbon is left off most comparisons.

“OpenAI Releases GPT 6.2 Astra & GPT 6.2 Sol, Beat Fable 5.2, Gemini 4 Carbon On Some Benchmarks” · Open the source

S3TestingCatalog17 Nov 2026neutral

Confirms availability, the Ultrafast rollout and the self-improvement rating, and embeds an early tester calling Astra uneven and harder to steer.

“OpenAI launches GPT-6.2 Astra and Sol on ChatGPT and APIs” · Open the source


Claims ✓0 ~0 ✗0 ?50 reports
First stamped 17 Nov 2026Last re-stamped 17 Nov 2026