Morph1k
Pulse/Almanac

Morph1k AI Almanac

A living index of the signals that mattered.

Pulse editions settle here beside era milestones — a curated index of tools, research, and the modern-AI turning points worth keeping.

Signals
61
Editions
04
Era
45
Videos
38
Entry type

View

Time lens
Editorial weight
61entries in focus

August 2026

13 entries
  1. A graphite grid of isolated agent sandboxes converging on a central reusable environment hub
    4 Aug 2026Open agentsShifts the rules

    Microsoft open-sourced the floor agents train on

    Orchard turns sandboxes, datasets, and evaluation into a reusable Kubernetes service — and ships recipes for coding, browser, and assistant agents.

    Open editionSource ↗
  2. Flowing voice waveforms crossing a live duplex stream with a separate async tool path
    4 Aug 2026OpenAIActually useful

    OpenAI rebuilt voice so it can stay live

    GPT-Live streams audio in both directions, delegates tool work on a separate path, and will underpin the upcoming GPT-Live API.

    Open editionSource ↗
  3. A 256-token canvas refining in parallel as diffusion noise settles into coherent text blocks
    4 Aug 2026GoogleWorth watching

    DiffusionGemma’s paper makes parallel text real

    Google’s July 31 technical report documents ~1,500 output tokens per second on an H100 by refining 256-token blocks in parallel — with a hybrid path back to autoregressive decoding.

    Open editionSource ↗
  4. A sealed frontier model vault with a transparent 30-day pre-release checkpoint gate
    4 Aug 2026AI policyShifts the rules

    Washington’s voluntary frontier-model tests are now on the calendar

    A June executive order’s 60-day deadline landed August 1. On August 4 OpenAI, Anthropic, Google, and Meta meet the White House on the rollout.

    Open editionSource ↗
  5. A monochrome synthetic portrait split across transparent registration layers
    2 Aug 2026AI policyShifts the rules

    AI labels just became enforceable in Europe

    Chatbots must identify themselves, deepfakes need visible labels, and synthetic media must carry machine-readable marks.

    Open editionSource ↗
  6. A dark isolated server chamber opening through a narrow breach to outside infrastructure
    2 Aug 2026AI safetyShifts the rules

    A test sandbox touched the real internet — and Claude kept going

    Anthropic found three incidents where cyber evaluations reached real organizations after a partner environment was misconfigured.

    Open editionSource ↗
  7. An aerial city map partly transformed into an imagined district beneath a closing dark veil
    2 Aug 2026Creative cultureWorth watching

    Google Earth learned to imagine places — then rolled it back

    A global Nano Banana feature turned real locations into generated concepts. Policy-breaking screenshots ended the experiment one day later.

    Open editionSource ↗
  8. An open graphite vault filled with model blocks and cold blue data paths
    2 Aug 2026Open modelsActually useful

    DeepSeek put its stronger coding agent behind MIT-licensed weights

    V4 Flash’s official build replaces the preview, adds faster speculative decoding, and publishes the 304B-parameter model under MIT.

    Open editionSource ↗
  9. A dark director’s table where contact sheets, film frames, audio, and motion diagrams converge
    2 Aug 2026Creative toolsActually useful

    Seedance 2.5 turns a pile of references into one directed scene

    One pass can now make a 30-second clip with sound, guided by up to 30 images, 10 videos, and 10 audio files.

    Open editionSource ↗
  10. Dark cinematic cover for the Claude Opus 5 story
    Video
    1 Aug 2026AnthropicShifts the rules

    Claude Opus 5 is the new everyday frontier

    Near Fable-level intelligence at half the cost — and fewer of the restrictions that made Fable awkward for normal use.

    Open editionSource ↗Watch ↗
  11. Atmospheric cover suggesting efficient model pipelines
    Video
    1 Aug 2026OpenAIActually useful

    GPT-5.6 just got dramatically cheaper to run

    After Sol / Terra / Luna launched, OpenAI passed efficiency gains into real price cuts — especially on Terra and Luna.

    Open editionSource ↗Watch ↗
  12. Cover suggesting deep mathematical reasoning
    1 Aug 2026ResearchWorth watching

    OpenAI’s next model already cracked ten open math problems

    An internal Astra build shipped Lean proofs for problems that sat open for a decade — for roughly $2,000 in compute.

    Open editionSource ↗
  13. Cover for open / local model story
    1 Aug 2026Open modelsActually useful

    July’s open models made private labs feel possible again

    From Moonshot’s giant Kimi K3 to fresh image stacks like FLUX 3 — the open floor moved up in one month.

    Open editionSource ↗

July 2026

03 entries
  1. Cover for the GPT-5.6 public launch story
    Video
    29 Jul 2026OpenAIShifts the rules

    GPT-5.6 Sol, Terra, and Luna went fully public

    After a limited government-reviewed preview, OpenAI’s three-tier frontier family landed in ChatGPT, Codex, and the API.

    Open editionSource ↗Watch ↗
  2. Cover for the Fable 5 return story
    29 Jul 2026AnthropicWorth watching

    Claude Fable 5 came back worldwide

    After the US export-control pause, Anthropic restored full public availability of Fable 5 — then spent July racing Opus 5 toward everyday use.

    Open editionSource ↗
  3. Cover for Meta Muse Spark story
    29 Jul 2026MetaActually useful

    Meta re-entered with Muse Spark and a paid API

    July stretched the frontier from three labs toward five — Meta’s Muse Spark 1.1 and first paid developer API were part of that shift.

    Open editionSource ↗

March 2026

04 entries
  1. Natural real-time voice waves flowing around a live camera conversation
    Video
    26 Mar 2026MilestoneReal-time voiceShifts the rules

    Gemini 3.1 Flash Live reached conversational speed

    Google’s strongest audio model paired natural rhythm, longer conversational memory, multilingual speech, and live vision across more than 200 countries.

    Modern AI eraSource ↗Watch ↗
  2. A reasoning conductor delegates parallel tasks to a constellation of small agents
    17 Mar 2026MilestoneSmall modelsShifts the rules

    GPT-5.4 mini and nano made agent teams economical

    Smaller reasoning models brought vision, tools, computer use, and skills to high-volume systems — and let a flagship delegate parallel work.

    Modern AI eraSource ↗
  3. A professional reasoning agent coordinating computers, tools, and applications
    Video
    5 Mar 2026MilestoneComputer useShifts the rules

    GPT-5.4 unified reasoning, coding, and computer use

    OpenAI moved Codex-grade agency into its mainline professional model, with a million-token context and search across large tool ecosystems.

    Modern AI eraSource ↗Watch ↗
  4. Compact intelligence moving at high speed through a vast volume of tasks
    3 Mar 2026MilestoneEfficient intelligenceActually useful

    Gemini 3.1 Flash-Lite priced intelligence for volume

    Google’s fastest and cheapest Gemini 3 model kept adjustable reasoning while targeting translation, moderation, interfaces, and simulations at scale.

    Modern AI eraSource ↗

February 2026

06 entries
  1. A high-speed visual production system combining world knowledge and precise composition
    26 Feb 2026MilestoneImage at scaleActually useful

    Nano Banana 2 made Pro-quality images fast

    Google compressed advanced world knowledge, editing, typography, and production control into a Flash image model built for mass use.

    Modern AI eraSource ↗
  2. A faceted multimodal reasoning core surrounded by audio, video, code, images, and data
    19 Feb 2026MilestoneCore reasoningShifts the rules

    Gemini 3.1 Pro made research-grade reasoning general

    Google moved the intelligence behind Deep Think into a broadly shipped multimodal baseline for complex problems.

    Modern AI eraSource ↗
  3. An AI research instrument exposing a hidden flaw among mathematical and physical systems
    Video
    12 Feb 2026MilestoneAI for scienceShifts the rules

    Gemini 3 Deep Think entered the research process

    A specialized reasoning update targeted messy, open-ended science — from hidden proof errors to physical systems and semiconductor growth.

    Modern AI eraSource ↗Watch ↗
  4. A meticulous AI planner carrying a vast archive through a sustained task
    Video
    5 Feb 2026MilestoneLong contextActually useful

    Claude Opus 4.6 learned to stay with the problem

    Anthropic’s flagship paired a one-million-token context window with more careful planning, stronger review, and longer autonomous runs.

    Modern AI eraSource ↗Watch ↗
  5. A human steering an autonomous engineering intelligence across a vast codebase
    5 Feb 2026MilestoneAgentic codingShifts the rules

    GPT-5.3-Codex expanded coding into computer work

    OpenAI fused frontier coding with professional reasoning in a steerable model built for research, tools, and long-running execution.

    Modern AI eraSource ↗
  6. A graphite command center coordinating parallel software agents and isolated worktrees
    Video
    2 Feb 2026MilestoneMulti-agent workShifts the rules

    The Codex app turned coding into agent orchestration

    A desktop command center made parallel agents, isolated worktrees, reusable skills, and scheduled automations part of one reviewable workflow.

    Modern AI eraSource ↗Watch ↗

December 2025

01 entries
  1. A professional knowledge engine coordinating science, documents, data, and long-running work
    11 Dec 2025MilestoneKnowledge workActually useful

    GPT-5.2 targeted the work behind the work

    Instant, Thinking, and Pro focused the frontier on spreadsheets, presentations, long documents, code, tools, science, and multi-step professional projects.

    Modern AI eraSource ↗

November 2025

03 entries
  1. An expert AI collaborator operating code, spreadsheets, slides, and computer interfaces
    Video
    24 Nov 2025MilestoneComputer useShifts the rules

    Claude Opus 4.5 made computer use practical

    Anthropic’s flagship sharpened coding, agents, and GUI interaction while cutting Opus pricing enough to move premium capability into daily work.

    Modern AI eraSource ↗Watch ↗
  2. A professional visual system of precise diagrams, maps, and complex composition
    Video
    20 Nov 2025MilestoneProfessional imageActually useful

    Nano Banana Pro gave image generation world knowledge

    Gemini 3 Pro Image combined visual reasoning, Search grounding, typography, and complex composition to move from playful edits toward professional graphics.

    Modern AI eraSource ↗Watch ↗
  3. A crystalline intelligence embedded across search, code, mobile, and dynamic interfaces
    Video
    18 Nov 2025MilestoneGoogleShifts the rules

    Gemini 3 shipped intelligence at Google scale

    Advanced reasoning, multimodality, agentic coding, and generative interfaces arrived across Gemini, Search, developer tools, and enterprise on day one.

    Modern AI eraSource ↗Watch ↗

October 2025

01 entries
  1. An intelligent browser window mapping context and actions across the web
    Video
    21 Oct 2025MilestoneAgentic browserShifts the rules

    ChatGPT Atlas put the agent inside the browser

    OpenAI rebuilt browsing around persistent page context, memory, and an agent that could act where the user already worked.

    Modern AI eraSource ↗Watch ↗

September 2025

01 entries
  1. A coherent multi-shot synthetic action film with synchronized sound
    30 Sept 2025MilestoneSynthetic videoShifts the rules

    Sora 2 turned generated video into a social medium

    Better physics, synchronized sound, persistent multi-shot worlds, and consent-based likeness cameos arrived inside a dedicated creation feed.

    Modern AI eraSource ↗

August 2025

02 entries
  1. One recognizable portrait preserved across a grid of imaginative edits
    Video
    26 Aug 2025MilestoneImage editingShifts the rules

    Nano Banana made identity survive the edit

    Google’s viral Gemini image model kept people and pets recognizable across scenes, combined references, and made precise edits through conversation.

    Modern AI eraSource ↗Watch ↗
  2. A unified crystalline core routing fast and deep intelligence pathways
    Video
    7 Aug 2025MilestoneUnified intelligenceShifts the rules

    GPT-5 ended the model-picker era

    A fast model, a deeper reasoning model, and a live router became one default system that decided how much thought each request deserved.

    Modern AI eraSource ↗Watch ↗

July 2025

01 entries
  1. An autonomous assistant orchestrating browser, terminal, calendar, and documents
    Video
    17 Jul 2025MilestoneConsumer agentsShifts the rules

    ChatGPT agent crossed from answers into actions

    Deep research, a visual browser, a terminal, files, and app connectors converged into one mode that could plan and execute multi-step work.

    Modern AI eraSource ↗Watch ↗

May 2025

02 entries
  1. Two advanced intelligence figures sustaining work across a deep archive
    Video
    22 May 2025MilestoneAgentic codingShifts the rules

    Claude 4 was built to stay with the task

    Opus 4 and Sonnet 4 pushed coding and long-running agent work forward, pairing hybrid reasoning with memory, tool use, and sustained focus.

    Modern AI eraSource ↗Watch ↗
  2. A cinematic generated world crossed by synchronized dialogue and ambient waveforms
    Video
    20 May 2025MilestoneGenerative mediaShifts the rules

    Veo 3 gave generated video a voice

    Google’s model generated image and synchronized sound together — ambient noise, effects, and spoken dialogue — collapsing another stage of production.

    Modern AI eraSource ↗Watch ↗

April 2025

01 entries
  1. A reasoning agent coordinating tools across a complex research board
    Video
    16 Apr 2025MilestoneTool reasoningShifts the rules

    OpenAI o3 learned to think with tools

    Reasoning expanded beyond internal deliberation: o3 could decide when to browse, run Python, inspect files, and reason directly over images.

    Modern AI eraSource ↗Watch ↗

March 2025

01 entries
  1. A conversational interface painting an original hand-drawn flying city
    Video
    25 Mar 2025MilestoneVisual cultureShifts the rules

    4o image generation turned the chat into a studio

    Reliable text, precise edits, remembered characters, and conversational iteration made image generation useful — then the Ghibli-style wave made it unavoidable.

    Modern AI eraSource ↗Watch ↗

February 2025

01 entries
  1. One intelligence monolith divided between instant and extended thought
    Video
    24 Feb 2025MilestoneHybrid reasoningActually useful

    Claude 3.7 put fast and deep thinking in one model

    Anthropic’s first hybrid reasoning model could answer instantly or think visibly for longer, while developers controlled the compute budget.

    Modern AI eraSource ↗Watch ↗

January 2025

01 entries
  1. A compact crystalline reasoning engine glowing in a deep blue trench
    20 Jan 2025MilestoneOpen reasoningShifts the rules

    DeepSeek-R1 broke the price of frontier reasoning

    An MIT-licensed Chinese reasoning model matched o1-class results on major benchmarks, exposed its training recipe, and shipped distilled versions small enough for local use.

    Modern AI eraSource ↗

September 2024

01 entries
  1. Luminous reasoning steps crossing a dark mathematical labyrinth
    Video
    12 Sept 2024MilestoneReasoningShifts the rules

    OpenAI o1 made thinking time a scaling dimension

    Instead of answering immediately, the new model family spent extra compute reasoning through a problem — trading speed for harder, more reliable work.

    Modern AI eraSource ↗Watch ↗

June 2024

01 entries
  1. A private intelligence core glowing across a premium device ecosystem
    Video
    10 Jun 2024MilestonePersonal AIShifts the rules

    Apple Intelligence brought AI into the operating system

    Apple framed generative models around personal context, actions across apps, on-device processing, and a new private cloud architecture.

    Modern AI eraSource ↗Watch ↗

May 2024

01 entries
  1. A human and synthetic figure joined by realtime voice and vision waves
    Video
    13 May 2024MilestoneRealtimeShifts the rules

    GPT-4o made conversation feel immediate

    One “omni” model handled audio, vision, and text with near-human response latency — laughing, interrupting, watching, and answering in real time.

    Modern AI eraSource ↗Watch ↗

March 2024

01 entries
  1. Three refined intelligence monoliths standing in a dark archive
    Video
    4 Mar 2024MilestoneAnthropicShifts the rules

    Claude 3 turned the frontier into a real contest

    Haiku, Sonnet, and Opus offered three deliberate balances of speed, cost, and intelligence — with Opus challenging GPT-4 at the top.

    Modern AI eraSource ↗Watch ↗

February 2024

01 entries
  1. A film strip dissolving into a generated nocturnal city
    Video
    15 Feb 2024MilestoneVideoShifts the rules

    Sora made generated video feel like cinema

    Minute-long, high-definition scenes with camera movement and visual continuity turned text-to-video from an experiment into a glimpse of synthetic filmmaking.

    Modern AI eraSource ↗Watch ↗

December 2023

01 entries
  1. Streams of text, image, audio, and code converging into a crystalline core
    Video
    6 Dec 2023MilestoneGoogleShifts the rules

    Gemini made multimodality Google’s foundation

    One model family spanning data centers and phones signaled that generative AI would flow through the entire Google product system.

    Modern AI eraSource ↗Watch ↗

September 2023

01 entries
  1. A conversational interface joined by an eye, camera, and voice waveform
    25 Sept 2023MilestoneMultimodalActually useful

    ChatGPT learned to see, hear, and speak

    Images and two-way voice moved the assistant beyond the text box and toward an ambient interface for the physical world.

    Modern AI eraSource ↗

July 2023

01 entries
  1. An open luminous gateway releasing model weights to a crowd of builders
    Video
    18 Jul 2023MilestoneOpen weightsShifts the rules

    Llama 2 gave open models commercial gravity

    Meta released capable pretrained and chat models for research and most commercial use, turning open weights into a credible platform strategy.

    Modern AI eraSource ↗Watch ↗

March 2023

02 entries
  1. A photoreal portrait emerging from a painterly synthetic dream
    Video
    15 Mar 2023MilestoneImageShifts the rules

    Midjourney V5 crossed the uncanny threshold

    Photorealism, wider style range, higher resolution — and hands that finally looked human — collapsed the distance between prompt and photograph.

    Modern AI eraSource ↗Watch ↗
  2. A crystalline multimodal intelligence prism joining text, images, and diagrams
    Video
    14 Mar 2023MilestoneFrontier modelShifts the rules

    GPT-4 raised the ceiling from fluent to formidable

    Stronger reasoning, professional-level benchmark results, and image understanding made the chatbot feel less like a novelty and more like infrastructure.

    Modern AI eraSource ↗Watch ↗

November 2022

01 entries
  1. A glowing conversational interface connected to a field of people
    Video
    30 Nov 2022MilestoneConsumer AIShifts the rules

    ChatGPT gave the AI boom an interface

    A free research preview wrapped a capable language model in dialogue — and made generative AI legible to hundreds of millions.

    Modern AI eraSource ↗Watch ↗

August 2022

01 entries
  1. A human eye and landscape resolving from a field of luminous noise
    Video
    22 Aug 2022MilestoneOpen weightsShifts the rules

    Stable Diffusion put image generation in public hands

    Downloadable weights moved text-to-image from a guarded cloud service onto personal hardware — and opened the floodgates to adaptation.

    Modern AI eraSource ↗Watch ↗

July 2022

01 entries
  1. A monumental fantasy citadel emerging from painterly mist
    13 Jul 2022MilestoneCultureShifts the rules

    Midjourney made AI imagery a public culture

    A Discord bot, a shared stream, and a recognizable aesthetic turned generation into a social creative ritual almost overnight.

    Modern AI eraSource ↗

April 2022

01 entries
  1. An astronaut riding a horse through a surreal museum landscape
    Video
    6 Apr 2022MilestoneImageShifts the rules

    DALL·E 2 made generated images look finished

    Four times the resolution, stronger composition, inpainting, and variations: text-to-image crossed from curiosity into a usable visual medium.

    Modern AI eraSource ↗Watch ↗

June 2021

01 entries
  1. Two luminous cursors collaborating inside a dark code editor
    Video
    29 Jun 2021MilestoneCodingShifts the rules

    GitHub Copilot put generative AI inside the editor

    Code stopped being only something a model could discuss. It became something the model could write beside you, line by line.

    Modern AI eraSource ↗Watch ↗

January 2021

01 entries
  1. Surreal gallery still of an avocado armchair suggesting early text-to-image culture
    Video
    5 Jan 2021MilestoneOpenAIShifts the rules

    DALL·E turned language into an image studio

    Text-to-image left the research demo and entered culture — avocado armchairs included — as generation became a creative verb.

    Modern AI eraSource ↗Watch ↗

May 2020

01 entries
  1. Nebula-like still suggesting vast model scale as a field of parameters
    Video
    28 May 2020MilestoneOpenAIShifts the rules

    GPT-3 made few-shot prompting a product primitive

    175 billion parameters, and a quieter revolution: ask in plain language, give a few examples, get useful work — without retraining.

    Modern AI eraSource ↗Watch ↗

February 2019

01 entries
  1. Cinematic still suggesting fluent machine-written language spreading outward
    Video
    14 Feb 2019MilestoneOpenAIShifts the rules

    GPT-2 made fluent generation feel unsettlingly real

    OpenAI’s staged release turned a research model into a culture event — the first time the public argued about AI writing at scale.

    Modern AI eraSource ↗Watch ↗

June 2018

01 entries
  1. Editorial still suggesting a model learning language before specializing
    Video
    11 Jun 2018MilestoneOpenAIActually useful

    GPT-1 proved unsupervised pre-training could transfer

    OpenAI’s first Generative Pre-trained Transformer showed a simple recipe: learn language first, then specialize — the seed of today’s assistants.

    Modern AI eraSource ↗Watch ↗

June 2017

01 entries
  1. Abstract dark still suggesting parallel attention across a sequence
    Video
    12 Jun 2017MilestoneArchitectureShifts the rules

    Attention Is All You Need rewired sequence models

    The Transformer dropped recurrence and made scale the new research language — the substrate every later GPT still stands on.

    Modern AI eraSource ↗Watch ↗

Morph1k Almanac

Built to grow with every Pulse cycle and every era milestone — a calm place to return when the feed moves on.

Back to latest Pulse ←