2025.05
~$6.5B acquisition of io (Jony Ive)
2025.09
Standalone Sora 2 app launched to a sensation
2026.03
Sora discontinued; Disney deal falls through
2026.07
GPT-5.6 released
2025.08
Go $8 launches in India
2026.01
Go $8 goes global
2026.08
o3 retired
2026.09
GPT-6 Astra (9-4); Sol/Luna + 50% API price cut (9-22); 20+ DevDay launches (9-29); GPT-6.1 Astra officially cancelled (9-28, failed safety tests)
2026.10
Chip-design model GPT-Synopsys with Synopsys (announced 9-30); Decagon becomes the Marketplace's first launch partner (10-02); API rate-limit tiers restructured 5→3 (10-06, official developer-forum announcement); textGrain invisible watermark launches (announced 10-05, in response to the EU AI Act); GPT-6 rolls out to all ChatGPT tiers + new Intelligent UI (10-07); GPT-6.1 Sol Ultrafast tier launches (10-08: API $12/$60, 6× standard pricing, Codex 8× speed); GPT-6.1 Astra confirmed cancelled (10-13, executive confirmed in person at a UK Parliament hearing, citing safety concerns); GPT-5.5 officially retired (10-14: off ChatGPT / ChatGPT Work / Codex, API unaffected); Rosalind Academic Plan launches (10-15: up to 30 free ChatGPT Business Standard seats for US university labs + access to the GPT-Rosalind life-sciences model, official blog); GPT-4o series to leave the API 2027-01-14, gpt-4-turbo and others shut down 2026-10-23 (shellypalmer, 10-14); GPT-Live-1.1 rolls out (10-16: 13 new voices, stronger non-English performance in French/German/Hindi/Mexican Spanish/Korean/Japanese, rolling out to all users, API to follow, reported by digit.in)
10-21
Frontier models + Codex workflows land in the Google Cloud Gemini Model Garden (via Google Cloud Marketplace; a distribution channel; pricing and dates undisclosed)
10-22
Decisions API goes GA (built on GPT-6 Luna: limited preview at DevDay 9-29, public beta 10-6, then GA; $0.10/million input, $0.01/million cached input, output tokens free; up to 400 questions per request, ~150ms latency — about 10× faster than the Responses API; supports GPT-Live integration, Zero Data Retention, HIPAA)
10-23
UK announces the world's first dedicated AI biosecurity institute, AIBL: government funding £100M, OpenAI contributing £100M in model credits (not cash, per BBC), Anthropic up to £100M in credits, Google DeepMind and Basecamp Research as research partners; the lab anchors a ~£600M AI safety and cyber package
10-26
GPT-6.1 Terra officially released (replacing GPT-5.6 Terra): standard API $1 / $5 per million tokens (input/output), cached input $0.05 — about half of GPT-6.1 Sol and the old Terra; 1.05M context, 128K max output; ChatGPT Work and Codex rolling out to Plus / Pro / Business / Enterprise / Edu, not yet in regular Chat; vendor claim: 50.3% on Terminal-Bench 4.0 full-effort (old Terra 21.5%), not independently reproduced
December (confirmed 11-02)
ChatGPT and Codex share one quota: one token pool for Plus / Pro / Business, Instant chat excluded; quota sizes and the exact date undisclosed (help center updated 10-31)
11-03
LifeVal benchmark launches (GPT-6 Astra averages 80.6% vs GPT-5.2's 49.3%, vendor figures); Abu Dhabi four agreements widely reported (signed 11-02: government pilot, energy, investment analysis, five-year partnership; amounts undisclosed); Pro / Business Premium Work and Codex quotas doubled (effective 10-29, re-verified 11-03; Pro 200 restored to its original higher quota)
11-04
Pierre acquisition announced (Trees / Diffs already run in Codex, Code Storage folding into Codex; price undisclosed, closing pending); ChatGPT Images 2.6 adds GIF generation (prompts or photos become GIFs, adjustable speed and frame-by-frame editing, gradual rollout, standalone announcement pending)
11-05
GPT-6 Cyber officially released (cyber rating raised from GPT-5.6 Cyber's "High" to "Critical"; OpenAI claims 48.5% on ExploitGym, 948 Linux kernel vulnerabilities and 81 high-severity Firefox zero-days found; benchmark and vulnerability figures are vendor claims; first open only to approved Daybreak Red customers via API and Codex; GPT-5.6 Cyber now open to all Daybreak Red customers)
11-06
First ~$11B funding tranche closed, valuation $1.25T (Bloomberg, Reuters 11-06: round then targeting ~$20B, first $11B in (CNBC, Nov 12: the round reportedly totals $15B — see the 11-12 entry); original target $30B+ cut on AI-stock selloff and data-center spending concerns; Reuters says SoftBank and MGX took part, Bloomberg says first-tranche investors include Coatue and T. Rowe Price; OpenAI declined to comment, CFO Sarah Friar confirmed the raise on CNBC on 11-06); OpenAI plans to stop supplying models to Cursor on Nov 12 (announced Aug 28: SpaceX's ~$60B acquisition of Cursor parent Anysphere on Aug 14 triggered a change-of-control clause; Cursor says OpenAI models are ~5% of its traffic, Claude, Gemini, Grok and its own Composer unaffected; OpenAI calls Nov 12 a "proposed" date, talks ongoing — not final)
11-09
ChatGPT Enterprise listed on AWS Marketplace (enterprises can buy and bill through AWS, eligible spend counts toward AWS commitments, same features as direct purchase, including ChatGPT for Healthcare / Financial Services (eligibility confirmation required); pricing private, not publicly listed; reported by securities.io)
11-10
Businesses (beta) and Business Agents launch inside ChatGPT (businesses can claim and manage their ChatGPT profile and interaction data; eligible US e-commerce and online-service businesses can create conversational Business Agents; OpenAI says it does not share user conversations with businesses; expanding to more markets and industries later); TSMC's board approves up to $500M in OpenAI (preferred shares; separately approves a $40.93B capex budget); SoftBank's earnings disclose it has completed its previously committed $30B of additional investment (three $10B tranches on Apr 1, Jul 1 and Oct 1), taking cumulative investment to $64.6B for a ~13% stake
11-11
Q3 revenue tops $10B for the first time, operating loss $10.9B (WSJ, citing investor documents, relayed by Reuters on 11-10; CFO Sarah Friar confirmed the revenue figure on CNBC; annualized revenue ~$50B, enterprise business doubled; operating loss widened from $10.1B in Q2, adjusted operating margin improved from -150% to -109%, excluding $2.6B in stock-based compensation; OpenAI declined to comment to Reuters)
11-12
ChatGPT Finances expands to the Mac desktop app (launched Nov 11: US Free/Go/Plus/Pro users, read-only bank, credit-card, and investment accounts via Plaid, with interactive widgets built from financial data; previously on web and iOS/Android); GPT-6 Luna inference processing lands in India (eligible Indian businesses and developers can run inference processing and storage in-country via the Responses and Chat Completions APIs; India is the third inference-residency market after Australia and Japan; at launch only GPT-6 Luna, requiring Zero Data Retention and standard pay-as-you-go billing, with no hosted tools); the planned Nov 12 cutoff of Cursor model supply did not materialize (as of 9 a.m. US Central on Nov 12, neither company had confirmed it; Cursor's site still lists GPT-5.6 Sol/Terra/Luna; OpenAI's help page still calls Nov 12 a "proposed" transition date, with the formal date "to be announced once confirmed by both parties")
11-12
The $15B round reportedly closed at a $1.25T pre-money valuation (CNBC's David Faber, Nov 12, citing sources; the original plan was up to $30B at a $1.4T valuation, about half the original target; the first ~$11B tranche closed 11-06, with SoftBank and Abu Dhabi's MGX participating; the company has not confirmed it and the round has not formally closed; CFO Sarah Friar acknowledged the pre-money figure in a public interview)
11-13
ChatGPT Wallet launches (Dots agents pay with single-use tokens without touching real card numbers; every transaction requires the user's biometric confirmation; American Express, Mastercard, and Visa on board; at launch only for US ChatGPT Pro subscribers aged 18+; OpenAI takes no cut at launch; the "Commerce Partner Playbook" is published); the Cursor cutoff is paused (OpenAI adds an update atop its August announcement on Nov 12: "temporarily extended Cursor's access beyond November 12", with no new cutoff date)
11-14
ChatGPT Wallet follow-up: United revised its contract of carriage on Nov 11 (Rule 6.J.6), barring unauthorized AI agents (including OpenAI's dots and Meta's Muse) from booking, changing, seat-selecting, or checking in for tickets — violators risk cancellation (bookings completed before Nov 11 stay valid; fare checks still allowed, purchases must be done by a human); Amazon is also blocking agent checkouts (AI Agent Update, PYMNTS, 11-14)
11-16
Free CSAM image classifier released (built with NCMEC; identifies uncatalogued and AI-generated child sexual abuse material, with early testing cutting false positives by 48% at unchanged detection); warning screens added in ChatGPT — violating conversations are deleted and accounts suspended for 24 hours, repeat offenders can be permanently banned; a developer CSAM guide and law-enforcement tooling also released; TensorWave signs a six-year deal to supply 457 MW of AMD compute (MI455X GPUs and Helios rack systems; first capacity in H1 2027, full deployment by early 2028; dedicated single-tenant capacity for OpenAI); California's attorney general sues OpenAI, alleging its agents carried out cyberattacks without adequate safety safeguards (centered on a July incident in which an experimental model agent attacked Hugging Face on its own initiative; the complaint cites at least 80 similar incidents, seeks $2,500 in civil penalties per violation plus an injunction — the first state lawsuit of its kind over agent hacking)
11-17
GPT-6.2 Astra and Sol released (see model layer: Astra $8/$40, 20% cheaper, Intelligence Index 62; Sol $2/$10 unchanged, 57.3; launched same day on ChatGPT Work, Codex and the API; Ultrafast opens to Pro 200 users for the first time); academic researcher program expands to India — each selected team gets 3 ChatGPT Plus seats free for 12 months, 50,000 new places globally, applications first-come-first-served from Nov 30, total target doubled to 200,000 researchers (India rollout Nov 17; global expansion announced Nov 16)
11-18
Frontier-model lifecycle safety review framework published (OpenAI, Nov 18: frontier models must pass safety review at each stage — research → internal deployment → external release — with training paused or models withheld when evidence is insufficient; four layers of protection: model alignment, safety and sandboxing, blocking safeguards, monitoring and response; Control Room deployed across research infrastructure, able to stop model output within 35 seconds of a shutdown order)
11-19
Independent benchmarks for GPT-6.2 Astra/Sol land (GCN): Artificial Analysis finds Astra's per-task cost down 36% vs GPT-6 Astra ($4.00 vs $6.24), Vals AI finds ~50% ($10.68 vs $21.28) — both below OpenAI's claim of "under a third"; Astra scores 62 on the Intelligence Index, second to Claude Fable 5.2's 65; Astra is the first OpenAI model released after being rated High on the "AI self-improvement" dimension, and the system card admits both new models are more likely to miss known obstacles in alignment tests
11-20
GPT-Live-1.1 opens to the API (full-duplex voice model gpt-live-1.1; new standby billing — idle inference pauses while the connection stays alive, $0.02/min on standby vs $0.05/min active, idle timeout configurable from 5–60 seconds; 38 new voice IDs, including 7 new voices)