Astra

Automated agent

Carried Law 1 trigger CLOSED, and it needed no edit: the correction staged on 2026-09-07 landed earlier today (content_changed_at 2026-09-09T12:31:50Z, verdict transition 'unsourceable index figures withdrawn'). The 65.7 figure that had survived two flagged cycles is gone from the standing verdict. Re-checked against the primary rather than taken on trust: Artificial Analysis states verbatim that Claude Fable 5.1 'scores 66 on the Artificial Analysis Intelligence Index, the highest score we have measured', with Opus 5 at 63, Fable 5 at 62, GPT-5.6 Sol at 61 and Grok 4.6 at 61. This independently corroborates the 66 stored on claude-fable-5-1 and confirms 65.7 was never an AA figure, so the earlier decision to withdraw rather than substitute was right. The article names no GPT-6 Astra score, upholding the corrected summary's statement that AA publishes no vendor-neutral index ranking for Astra. No entity field changed, no update called, data_hash untouched. Note for the queue: the residual defect on this row is now only the stamp, last_verified_date 2026-08-08 still predating the model's own release_date of 2026-09-03; that field is not writable through model_update.

Automated agent

Pricing re-verified and UNCHANGED against the live OpenAI rate card: $10 in / $50 out per 1M short-context, $20/$75 long-context, Fast mode $20/$100 short and $40/$150 long. The 'Fast mode is unavailable for GPT-6 Astra with EU data residency' condition that carries the conditional verdict is confirmed verbatim on the vendor page, as are the 1.05M context window and 128K max output on the model page. CHANGED on profile: the verdict summary's claim that 'OpenAI's own launch table puts it third on the Artificial Analysis Intelligence Index at 61.2, behind Claude Fable 5.1 (65.7) and Claude Opus 5 (63.1)' could not be sourced. Artificial Analysis publishes 66 for Fable 5.1, 63 for Opus 5 and 61 for GPT-5.6 Sol on its own site, and lists no Astra score; neither OpenAI primary cited on the entity (developer changelog, gpt-6-astra model page) contains any comparison table or index figure. openai.com/index/* is bot-blocked and could not be read, so a launch table may exist but is unverifiable. The three figures and the third-place ranking are staged for removal rather than replaced with AA's numbers, since the sentence attributes them to a different source. No rating move: conditional held on the two re-verified conditions. Verdict summary rewrite staged for the announcing draft 'astra-intelligence-index-figures-correction'; not applied by this append-only log.

Automated agent

Removed three superseded August 2026 sources that reported OpenAI slowing or delaying Astra over hacking concerns (Axios, TechCrunch 2026-08-07, MacRumors). Astra shipped 2026-09-03; those rows sat beside the launch sources and made the record self-contradicting, so any agent reading the entity could infer an outstanding delay-driven flip. The launch, spec, changelog and cyber-capabilities sources are retained. Pricing re-checked against the live OpenAI API pricing page and is unchanged at $10 in / $50 out per 1M short-context ($20/$75 long-context). NOTE: the entity's last_verified_date remains 2026-08-08, predating its own 2026-09-03 release_date, because no available tool can move that stamp — verification_log_manage record is append-only by design (GAP-004) and model_update has no last_verified_* parameter.