Permanent copy of issue 66. Today's issue is always the front page.
Previous issue | This week | Next issue | archive
Aleph Alpha said Monday it has released Kolibri, an open-weight model with 78 billion parameters that a customer can run on machines the customer controls. Google's help page says that from October 9 a free Gemini account keeps only Flash-Lite. Deutsche Telekom said AI and automation should cut about 2.5 billion euros in indirect costs by 2030, compared with 2023. Sam Altman told Politico the public should accept some harm so people can use the technology. A customer can keep the file. A free account cannot keep the stronger model.
Date: 2026-10-05
Updated: 2026-10-05 06:10
Status: LIVE
Broadcast
Aleph Alpha put Kolibri on Hugging Face under Apache 2.0, and its card says the file needs about 78 gigabytes of memory. Google marks Flash-Lite for an unpaid account and leaves Flash and Pro unmarked. Telekom says agents in the United States now handle 40 percent of customer contacts. The weights are a file. The phone call is still the company's.
Watch
Stories
Aleph Alpha said Monday it has released Kolibri, a model with 78 billion parameters and about 3 billion active on each token, under an Apache 2.0 license. The Heidelberg company said the weights have been available since October 3 and are meant to run on infrastructure the customer controls.
-- Aleph Alpha's card says the file needs about 78 gigabytes of memory. That is a server, not a laptop.
2026-10-05 / Aleph Alpha / chip, efficiency / local / #aleph-alpha-releases-kolibri-weights
Google's support page says that from October 9, a personal Gemini account without a paid plan keeps Flash-Lite. The same table marks Flash and Pro for higher plans and does not mark them for an account without a plan.
-- A person who does not pay keeps the light model. The stronger answers sit behind a plan.
2026-10-05 / Google / efficiency / central / #google-limits-free-gemini-to-flash-lite
Deutsche Telekom said Monday that AI and automation should cut about 2.5 billion euros in indirect costs by 2030, compared with 2023. The company said that in the United States, customer-service calls have fallen 55 percent and that AI agents now handle 40 percent of customer contacts.
-- The 2.5 billion euros is a target against 2023, not a lower phone bill this month.
2026-10-05 / Deutsche Telekom / energy / central / #deutsche-telekom-targets-2-5-billion-euros-in-cu
OpenAI chief executive Sam Altman told Politico on Sunday that the world should accept some bad things happening so people can use the technology. He said it would be unacceptable for one laboratory in San Francisco to hold the model and decide who receives the benefits.
-- He is arguing for access to a product OpenAI still runs. He did not publish a file.
2026-10-04 / SiliconANGLE / chip / mixed / #altman-says-the-public-should-accept-some-harm
Axis
local [#########################-----] central
Score: 82
clusters still winning
unchanged
Aleph Alpha released Kolibri, 78 billion parameters with about 3 billion active, under a license that lets a customer keep the weights. Google's free Gemini account keeps Flash-Lite from October 9 and drops Flash and Pro. Deutsche Telekom set a 2.5 billion euro cost target for 2030 against 2023. Altman argued for lighter rules and against one lab holding the model. The cluster still holds the training chips, the free meter, and the phone network.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
92 | mixed | No new chip order landed. Aleph Alpha's card says Kolibri's pre-training used 768 Nvidia B200 chips for 21 days. That run is finished. | ||
| ENERGY |
|
99 | central | No new power contract landed. Deutsche Telekom said it will put part of its 2027 savings into Germany's fiber network, not into a new plant. | ||
| EFFICIENCY |
|
57 | mixed | Kolibri uses about 3 billion of 78 billion parameters on each token, Aleph Alpha said. Google's unpaid Gemini account keeps Flash-Lite and drops Flash and Pro on October 9. |
Line charts (0=local 100=central 2026-09-20 to 2026-10-05)
Market overlay (2026-09-06 to 2026-10-05)
Latest: BTC $86,044 / hashrate ~959.55 EH/s / token markers avg $5.3033/M / axis 82. These lines are different units. We compare timing, not dollars-to-hashrate.
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 $86,044 / hashrate ~960 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)
| # | Model | Org | CC | Weights | Tier | In | Out | Blend |
| 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 |
If useful, support the broadcast.
bc1qfs3yw5qlq8sxs50crzh3g84ug27gvww9npu85n
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