$ tail -f /var/log/acceleration The AI acceleration timeline — frontier model releases, dated and sourced

Every frontier model release, compute buildout and culture moment since the movement got its name — newest first, every entry sourced. Updated weekly. The live counters on the homepage run on this same data, and the token cost calculator prices the models listed here.

$ uptime --frontier Days since the last frontier AI releases

$ tail -n 28 /var/log/acceleration The full AI acceleration log, 2022–2026

1. 2026-09-01 MODEL

   Claude Fable 5.1 ships

   Anthropic's Mythos-class update lands: Terminal-Bench-Science more than doubles to 52.6%, cache reads drop 75% to $0.25 per Mtok, base price holds at $10 / $50. Mythos 5.1 stays gated to vetted US orgs under Project Glasswing.

2. 2026-08-17 COMPUTE

   Nvidia underwrites 4.25 GW for OpenAI in Ohio

   Nvidia backs up to $105B of financing for OpenAI's Pike County campus — 4.25 GW in the first phase, 3.75 GW optional, online from 2028. SoftBank's SB Energy builds it and pledges 10 GW of new generation. The single largest compute financing on record.

3. 2026-08-12 MODEL

   Grok 4.6 ships

   xAI's new flagship matches GPT-5.6 Sol on the Artificial Analysis index, one point behind Fable 5, at $2 / $6 per Mtok with a 500K context. Live day one in Cursor, Grok Build and the API. Grok 5 is still in training.

4. 2026-08-10 POLICY

   GPT-5.6 Cyber ships to vetted partners only

   OpenAI releases a cyber-tuned Sol variant to a trusted tier — Accenture, IBM, CrowdStrike, Cloudflare — through its Daybreak Red program. Frontier capability now ships by clearance level, not by API key.

5. 2026-08-03 MODEL

   Qwen3.8-Max ships

   Alibaba's 2.4T-parameter MoE (95B active) lands at $2 / $6 per Mtok, with the weights following on Hugging Face ten days later — the first Max-class Qwen ever released openly. Two open '2T+' models from China inside a month.

6. 2026-08-01 MODEL

   OpenAI unveils Astra with ten solved open problems

   An internal build of OpenAI's next model family closes ten math and TCS problems open for a decade or more — for about $2,000 of inference. Altman demos the multi-agent system to senators and the White House economic team; no release date, and no decision yet on calling it GPT-6.

7. 2026-07-21 MODEL

   Google ships three Flash models, still no 3.5 Pro

   Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber land, but the Pro promised at I/O in May stays unshipped after 'internal delays'. Almost as a footnote: DeepMind has started its most ambitious pre-training run yet, for Gemini 4.

8. 2026-07-16 MODEL

   Kimi K3 ships

   Moonshot's 2.8-trillion-parameter open-weight model lands — the first open '3T-class' model, trailing only Fable 5 and GPT-5.6 on overall capability. China closes the frontier gap.

9. 2026-07-09 MODEL

   GPT-5.6 opens to the public

   After a two-week federal review, OpenAI clears GPT-5.6 (Sol, Terra, Luna) for everyone — shipped alongside ChatGPT Work, an agent built to carry out whole jobs. The gate lifts.

10. 2026-07-08 MODEL

    Grok 4.5 ships

    xAI's V9 coding model lands as Grok 4.5 — 1.5T parameters trained on Cursor data, priced at $2 / $6 per Mtok. Still not the fabled 6T Grok 5, which keeps slipping.

11. 2026-06-30 MODEL

    Claude Sonnet 5 ships

    Anthropic's most agentic Sonnet becomes the default model — near-Opus performance at Sonnet speed and price, aimed squarely at running agents cheaply.

12. 2026-06-26 POLICY

    GPT-5.6 previews under federal gate

    OpenAI unveils GPT-5.6 — Sol, Terra and Luna — but access is limited to ~20 US-government-vetted partners pending a pre-release review: the frontier's first regulatory chokepoint.

13. 2026-06-09 MODEL

    Claude Fable 5 ships

    Anthropic opens the Claude 5 family: always-on adaptive thinking and a 1M-token context window.

14. 2026-04-23 MODEL

    GPT-5.5 ships

    OpenAI's flagship update lands in ChatGPT and Codex; API access follows a day later.

15. 2025-11-24 MODEL

    Claude Opus 4.5 ships

    Anthropic's strongest coding model to date lands days after Gemini 3 — the frontier cadence compresses to weeks.

16. 2025-11-18 MODEL

    Gemini 3 ships

    Google's frontier line jumps again; agentic coding and long-horizon tasks improve another step.

17. 2025-11-12 MODEL

    GPT-5.1 ships

    OpenAI iterates the GPT-5 line with adaptive reasoning effort and faster default paths.

18. 2025-08-07 MODEL

    GPT-5 ships

    OpenAI unifies reasoning and fast paths behind one router; frontier access goes mass-market.

19. 2025-05-22 MODEL

    Claude 4 ships

    Anthropic's Opus 4 and Sonnet 4 push sustained autonomous coding into production use.

20. 2025-01-21 COMPUTE

    Stargate announced: $500B

    SoftBank, OpenAI, Oracle and MGX commit $500B over four years — $100B deploying immediately — for US AI datacenters.

21. 2025-01-20 MODEL

    DeepSeek-R1 open weights

    Frontier-class reasoning released under MIT license at a fraction of incumbent training cost.

22. 2024-09-12 MODEL

    OpenAI o1: the reasoning era

    Test-time compute becomes the new scaling axis — models that think before they answer.

23. 2024-05-13 MODEL

    GPT-4o: omni goes realtime

    Native audio-vision-text at half the price of GPT-4 Turbo. Frontier intelligence keeps getting cheaper.

24. 2024-03-04 MODEL

    Claude 3 ships

    Opus matches GPT-4; for the first time the frontier is contested by more than one lab.

25. 2023-10-16 CULTURE

    The Techno-Optimist Manifesto

    Marc Andreessen publishes the manifesto; e/acc ideas enter mainstream venture discourse.

26. 2023-03-14 MODEL

    GPT-4 ships

    The capability jump that reset every timeline. $30 per million input tokens at launch.

27. 2022-11-30 MODEL

    ChatGPT launches

    100M users in two months. The acceleration becomes visible to everyone.

28. 2022-05-31 CULTURE

    e/acc is born

    A pseudonymous newsletter by @zestular, @BasedBeffJezos, @bayeslord and co. names the movement: build, accelerate, ascend the Kardashev gradient.