Underclassic   rss   archive

Issue 58: OpenAI Pauses Training of Latest Models

Permanent copy of issue 58. Today's issue is always the front page.

Previous issue  |  This week  |  Next issue  |  archive

OpenAI said it has paused training of its latest models and will restart only when it has additional safeguards, the Associated Press reported. Australia's Senate asked the chiefs of OpenAI and Anthropic to appear at an inquiry into AI and data centers. The United States and China said they will open a channel for AI-related incidents. TechCrunch reported that Google is testing a buy button inside Gemini for Flipkart shoppers in India.

Date: 2026-09-27
Updated: 2026-09-27 06:07
Status: LIVE


Broadcast

Anthony Albanese said Saturday in Sydney that the agent cases showed humans needed to stay in charge. The Senate asked Sam Altman and Dario Amodei to appear. A person who wants Claude or ChatGPT this morning still rents the account. Neither summons hands over a file.

Watch


Stories

OpenAI pauses training of its latest models

The Associated Press reported that OpenAI has paused training of its latest models and will restart only when it is confident it has additional safeguards. The company said it expects to halt training again as new problems appear, and this is the second pause in three months.
-- A subscriber cannot restart that training run. The switch sits with OpenAI.
2026-09-26 / The Guardian / efficiency / central / #openai-pauses-training-of-its-latest-models

Australia calls OpenAI and Anthropic chiefs to inquiry

The Guardian reported Saturday that Australia's Senate asked Sam Altman and Dario Amodei to appear before a Greens-led inquiry into AI and data centers. Prime Minister Anthony Albanese said the agent cases showed that humans needed to stay in charge, and Environment Minister Murray Watt called OpenAI's conduct completely unacceptable.
-- The inquiry can summon the companies. It cannot hand a department a copy of the model.
2026-09-26 / The Guardian / efficiency / central / #australia-calls-openai-and-anthropic-chiefs-to-i

US and China agree an AI incident channel

China and the United States said Saturday they will open a channel for AI-related incidents and hold a dialogue in November on the technology's risks and benefits, the Boston Globe reported. The same statements extended a tariff truce by two months and kept a pledge to ease tariffs on about $30 billion of goods.
-- The channel is a line between two capitals. It does not put a model on a household computer.
2026-09-26 / Boston Globe / efficiency / mixed / #us-and-china-agree-an-ai-incident-channel

Google tests a buy button inside Gemini

TechCrunch reported Saturday that Google is testing a buy button on some Flipkart listings inside Gemini and AI Mode, so a shopper in India can reach checkout without leaving the chat. Google did not confirm the test, and a spokesperson said only that the company is always testing ways for people to connect with businesses.
-- If the test holds, the purchase stays inside Google's chat rather than in a program the shopper runs.
2026-09-26 / TechCrunch / efficiency / central / #google-tests-a-buy-button-inside-gemini


Axis
local [#######################-------] central
Score: 77   clusters still winning   up 1
OpenAI stopped training of its latest models and kept the restart decision. Australia asked the lab chiefs to a Senate inquiry. Washington and Beijing agreed a channel for AI incidents. The cluster still holds the model.

Forces

ForceMeterScorePullNote
CHIP / CAPEX
..
90 central No new chip a household can buy landed on Sunday's wire. The training pause runs on OpenAI's machines.
ENERGY
..
95 central No new power contract was confirmed today. The labs that rent out the models still pay the cluster bill.
EFFICIENCY
..
56 central OpenAI can stop its own training run. A subscriber cannot. Google's reported checkout, if the test holds, stays inside Gemini.

Line charts (0=local 100=central 2026-09-12 to 2026-09-27)

Axis score line chart

Forces line chart

Market overlay (2026-08-29 to 2026-09-27)

Latest: BTC $84,915 / hashrate ~951.37 EH/s / token markers avg $5.3033/M / axis 77. These lines are different units. We compare timing, not dollars-to-hashrate.

Market overlay line chart

Not shown: GPU cloud $/hour, hub electricity, miner hashprice, or lab cost per token -- none of those are a clean public feed.

Power and list prices
BTC $84,915 / hashrate ~951 EH/s. List-price gap (not a subsidy %): 44.4x list gap (blended) GPT-5.6 Luna -> Claude Fable 5 ($0.45/M to $20.00/M). (mempool.space + Coinbase/CoinGecko)

Marker prices (USD per 1M tokens; verified 2026-08-19; blended = (3*input + 1*output) / 4; -- = open/self-host or no single public list)

#ModelOrgCCWeightsTierInOutBlend
1 Claude Sonnet 5 Anthropic US closed
undisclosed
workhorse $2.00 $10.00 $4.00
2 GPT-5.5 OpenAI US closed
undisclosed
flagship $5.00 $30.00 $11.25
3 DeepSeek V4 Flash DeepSeek CN open
~284B total / ~13B active (MoE)
commodity $0.44 $1.32 $0.66

Full 20-model board


If useful, support the broadcast.
bc1qfs3yw5qlq8sxs50crzh3g84ug27gvww9npu85n
BTC QR

Subscribe

$ echo you@email.com | underclassic --subscribe

One email per issue. No tracking pixels, no filler.

home  |  rss  |  page 2  |  page 3  |  archive  |  models  |  subscribe  |  data  |  board  |  mcp  |  llms.txt