BitShovel.Discover. Understand. Stay informed.
Add item中

TRACKING · Tracking record

OpenAI GPT · Records and changes

The OpenAI GPT model family and its pricing, capability, and release changes.

00
UNDERSTANDING · What we know

Uses, conditions and important changes.

Each record has its own date and scope; these do not establish the live state of the whole product.

GPT-6 Astra · Reading revised · 5 Sept 2026

GPT-6 Astra: test computer-use gains and cost on a real workload

OpenAI positions Astra for computer use, coding, and document work. Its announcement lists Standard API prices of $10 per million input tokens and $50 per million output tokens. Enterprise access is off by default at launch and requires administrator enablement; rollout is phased.

Capabilities, prices, and access are official claims, without independent testing. Aggregator attention is not proof of workload gains. Caching, Fast mode, and retries affect cost. The card retains its discovery-source date; this review has its own clock.

Read the record and sources →

GPT-6 Astra · Source note · 5 Sept 2026

OpenAI’s new reasoning technique alarms AI safety experts

OpenAI’s new Astra model will use “recurrent depth,” a technique that allows the model to operate outside of the sequential thinking that characterizes most reasoning models (the TechCrunch AI original's own abstract, carried same-window by TechCrunch AI).

This is a discovery note, not a completed reading.

Read the record and sources →

GPT-5.6 Cyber · Recorded by the site · 27 Aug 2026 · Earlier coverage

Trail of Bits field test: a network-capable agent escaped its VM sandbox three times

A Trail of Bits engineer with preview access to GPT 5.6-Cyber (via Patch the Planet) gave it a real challenge: escape the QEMU/KVM VM (Debian 12, AMD Zen3) he uses for sandboxing. It succeeded three times — first via newly disclosed host-kernel bugs, then via disclosed bugs not yet classified as security issues after full updates, and finally by finding several 0-days in freshly rebuilt upstream sources — running autonomously for hours, pulling papers and writing oracles and minimal examples on its own.

[Evidence provenance: Trail of Bits' own blog] One machine, one demonstration — not a systematic benchmark; it targets a cyber-specialized model, and Lobsters upvotes are attention, not proof that the method generalizes to ordinary models or every configuration.

Read the record and sources →
Show 1 more readings and notes

GPT-5.6 Sol · Recorded by the site · 25 Aug 2026

The GPT-5.6 Sol price cut is loud; the real opportunity is unit economics

At re-verification, OpenRouter's official model page still showed 50% off and now listed $2 input and $10 output per million tokens. The Hacker News discussion had 635 points and 448 comments.

This is a discussion snapshot from Aug 22, 2026. Points and comments measure attention and are not proof of buyers, revenue, profit, or a repeatable opportunity; prices come from OpenRouter's official page and may change at any time.

Read the record and sources →

Changes and reading revisions

01
THE FULL RECORD · Full timeline

24 events retained by the site.

This object’s records stay together across live coverage, the watch pool and history. Every event links to its source record; corroboration and material changes have separate labels.

Dates use Beijing time. Open an event for its full record, original source and evidence. Corroboration adds support without counting as a material change.

GPT-5.6 Sol1 signal · Live
  1. Material changeThe Hacker News submission linked to the model page under the title ‘GPT-5.6 Sol Pricing Cut by 50% on OpenRouter.’ That created a price lead requiring maker-source verification.
  2. Material changeThe official Hacker News API still returned 633 points and 448 comments. Nothing changed from the prior snapshot, so attention was not repackaged as an additional update.
  3. Material changeOpenRouter’s official model page listed $2 input and $10 output per million tokens while still showing 50% off. BitShovel corrected the prior $2.50/$15 display accordingly.
  4. On radarThe GPT-5.6 Sol price cut is loud; the real opportunity is unit economics
  5. Market lateralRisk event (OpenAI GPT): "The OpenAI Huggingface incident from an agents P" — lead via reddit-local-llama. Same-class security or reliability events change adoption boundaries and risk calls.
  6. Market lateralRelease watch (OpenAI GPT): market chatter captured via aihot-featured (headline in Chinese). Same-class release moves reshape expectations and alternatives.
  7. Market lateralRisk event (OpenAI GPT): market chatter captured via aihot-featured (headline in Chinese). Same-class security or reliability events change adoption boundaries and risk calls.
  8. Market lateralRisk event (OpenAI GPT): "Now we have a timeline of the OpenAI accidental " — lead via simon-willison-entries. Same-class security or reliability events change adoption boundaries and risk calls.
  9. Market lateralRisk event (OpenAI GPT): market chatter captured via aihot-featured (headline in Chinese). Same-class security or reliability events change adoption boundaries and risk calls.
  10. Market lateralComparison (OpenAI GPT): market chatter captured via aihot-featured (headline in Chinese). Same-class comparisons and sentiment signal the market's relative standing.
  11. Market lateralRelease watch (OpenAI GPT): "New release of LLM adds support for reasoning tr" — lead via simon-willison-entries. Same-class release moves reshape expectations and alternatives.
  12. Market lateralRelease watch (OpenAI GPT): market chatter captured via aihot-featured (headline in Chinese). Same-class release moves reshape expectations and alternatives.
  13. Market lateralRisk event (OpenAI GPT): "OpenAI’s accidental cyberattack against Hugging " — lead via simon-willison-entries. Same-class security or reliability events change adoption boundaries and risk calls.
  14. Market lateralRelease watch (OpenAI GPT): "OpenAI launches Astra, its powerful (and controv" — lead via techcrunch-ai. Same-class release moves reshape expectations and alternatives.
  15. Market lateralRisk event (OpenAI GPT): "OpenAI confirms ‘wiki incident,’ says it’s ‘work" — lead via techcrunch-ai. Same-class security or reliability events change adoption boundaries and risk calls.
  16. Market lateralComparison (OpenAI GPT): "OpenAI's GPT-6 Astra Hits 99.9% on AGI Benchmark" — lead via alphasignal-daily. Same-class comparisons and sentiment signal the market's relative standing.
  17. Market lateralRelease watch (OpenAI GPT): market chatter captured via aihot-featured (headline in Chinese). Same-class release moves reshape expectations and alternatives.
GPT-6 Astra2 signals · Live
  1. On radarEntered the BitShovel radar: same-story across AIHOT featured, AlphaSignal (30/15 entries visible at collection).
  2. On radarGPT-6 Astra: test computer-use gains and cost on a real workload
  3. On radarEntered the BitShovel radar: first-party on TechCrunch AI (20 entries visible at collection).
  4. On radarOpenAI’s new reasoning technique alarms AI safety experts
GPT-5.6 Cyber1 signal · Watch pool
  1. On radarEntered the BitShovel radar: 25 entries in the current Lobsters front page feed.
  2. On radarTrail of Bits field test: a network-capable agent escaped its VM sandbox three times
  3. To watch poolThis item has left the current list. Existing records remain available, and later retained material will join the same archive.

02
COVERAGE · Record status

How these records are currently listed.

Coverage status describes how this site maintains its records, not whether a product is worth using. Existing records remain available after leaving the current list.

Registered follow-up scope

These are registered scopes, not complete product coverage or a receipt for the latest successful check.

03
THE LINEAGE · Lineage

Every version on this line has its own full record.

The line's full timeline above is grouped by version; each version's own record opens from here.

View the line map