2026-09-27 notis CATEGORY Show Read as Find More AboutArchiveCorrectionsThe numbers
NOTICE OF THE DAY
September 2026 3 NOTICES
MTWTFSS
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

Marked days carry a notice. The archive opens in the register.

The whole week ›
11:44
0/3
WEATHER SKY STACK

Take the day one card at a time

today

My page

Everything you saved, what you asked to be told about, and how often we write. Kept in this browser, not on an account.

SAVED FOLLOWING COLLECTIONS READING SINCE
Nothing saved yet. The bookmark on any card keeps it here, and the notices you save are what propose the features in the monthly issue. read today
Pricing watch Anything that moves a published rate.
Migration risks Deprecations and successor gaps.
One notice a day, the morning it is checked. Five a week.The week's five in one email, Friday at 07:00.Nothing daily. Just the monthly issue when it publishes.
Models 27 SEP

Supersonic Labs releases Julia-1, a 144.3M-parameter decision model with a 550.5 MiB checkpoint

Apache licenceJuliaSupersonic LabsmmBERT
Quoted from huggingface.co on 2026-09-27: At 144.3M parameters, it turns a state, a question, and possible answers… Quoted from huggingface.co, checked 2026-09-27. “ At 144.3M parameters, it turns a state, a question, and possible answers into one clear decision.” quoted from huggingface.co · checked 2026-09-27

Built on mmBERT-small, the Apache-2.0-licensed classifier takes a state, a question and up to 20 options, and returns one of them.

SOURCEHugging Face (Supersonic Labs model card) · 1 source
CHECKED27 Sep, 04:30
Models read
Quoted from huggingface.co on 2026-09-27: At 144.3M parameters, it turns a state, a question, and possible answers… Quoted from huggingface.co, checked 2026-09-27. “ At 144.3M parameters, it turns a state, a question, and possible answers into one clear decision.” quoted from huggingface.co · checked 2026-09-27 Models / 27 SEP

Supersonic Labs releases Julia-1, a 144.3M-parameter decision model with a 550.5 MiB checkpoint

Supersonic Labs released Julia-1, a 144.3-million-parameter open-weights model for classification and binary judgments rather than text generation. Built on JHU CLSPs mmBERT-small, it accepts a context, a question, and 2 to 20 candidate answers, and its FP32 weights occupy 550.5 MiB. The model artifacts are licensed under Apache 2.0, and the checkpoint evaluation, dated September 24, 2026, is the first release in what the company calls the Julia family of decision models.

TAGSApache licenceJuliaSupersonic LabsmmBERT
Tooling 27 SEP

Claude Code 2.1.280 sets Opus 5.5 as the default Opus model, with a 1M-token window

AnthropicClaudeClaude Code
Quoted from code.claude.com on 2026-09-27: Added Claude Opus 5.5 ( claude-opus-5-5 ), now the default Opus model… Quoted from code.claude.com, checked 2026-09-27. “ Added Claude Opus 5.5 ( claude-opus-5-5 ), now the default Opus model — 1M context, $4/$20 per Mtok with $0.20/Mtok cache reads” quoted from code.claude.com · checked 2026-09-27

The changelog lists Opus 5.5 at 4 dollars per million input tokens and 20 dollars per million output tokens.

SOURCEAnthropic · 1 source
CHECKED27 Sep, 04:43
Tooling read
Quoted from code.claude.com on 2026-09-27: Added Claude Opus 5.5 ( claude-opus-5-5 ), now the default Opus model… Quoted from code.claude.com, checked 2026-09-27. “ Added Claude Opus 5.5 ( claude-opus-5-5 ), now the default Opus model — 1M context, $4/$20 per Mtok with $0.20/Mtok cache reads” quoted from code.claude.com · checked 2026-09-27 Tooling / 27 SEP

Claude Code 2.1.280 sets Opus 5.5 as the default Opus model, with a 1M-token window

The Claude Code changelog from Anthropic lists 114 entries under version 2.1.280, dated September 22, 2026. It fixes writes through a symlinked path being judged by their in-tree spelling: the permission prompt now names where the write actually lands, and acceptEdits, allow rules and auto mode no longer approve a write that lands outside the project tree. The same release adds Claude Opus 5.5 as the default Opus model, with a 1M-token context window priced at 4 dollars/20 dollars per million input/output tokens and 20 cents per million tokens for cache reads.

TAGSAnthropicClaudeClaude Code
Anthropic · 1 source tap to close
Research 27 SEP

DeepSeek describes DSec, a sandbox platform for agentic training handling 3 million sandboxes a day

3FSDSecDeepSeekSandboxing
Quoted from arxiv.org on 2026-09-27: A single production-scale unit of DSec spans around 160 nodes, serving about… Quoted from arxiv.org, checked 2026-09-27. “ A single production-scale unit of DSec spans around 160 nodes, serving about 3 million sandboxes per day; in production, it supports over 380,000 concurrent sandboxes and sustains over 5,000 sandbox creations per second.” quoted from arxiv.org · checked 2026-09-27

The paper details production infrastructure for agentic training, spanning container, microVM and full-VM backends behind one SDK.

SOURCEarXiv (DeepSeek) · 1 source
CHECKED27 Sep, 04:30
Research read
Quoted from arxiv.org on 2026-09-27: A single production-scale unit of DSec spans around 160 nodes, serving about… Quoted from arxiv.org, checked 2026-09-27. “ A single production-scale unit of DSec spans around 160 nodes, serving about 3 million sandboxes per day; in production, it supports over 380,000 concurrent sandboxes and sustains over 5,000 sandbox creations per second.” quoted from arxiv.org · checked 2026-09-27 Research / 27 SEP

DeepSeek describes DSec, a sandbox platform for agentic training handling 3 million sandboxes a day

DeepSeek and collaborators published a paper on DeepSeek Elastic Compute (DSec), a sandbox infrastructure built for large-scale agentic training, describing a unified SDK over FnCall, container, microVM and full-VM execution backends plus a distributed filesystem called Fire-Flyer File System (3FS). The paper reports the production system handles about 3 million sandboxes daily, supports more than 380,000 concurrent sandboxes, sustains over 5,000 sandbox creations per second, and that a single production unit spans around 160 nodes. The paper was submitted to arXiv on September 19, 2026.

TAGS3FSDSecDeepSeekSandboxing
NOTHING IN THIS FILTER No notice today matches. The archive goes back further. open the archive →
That is the day. Three notices, all opened at source. Short of the target: three today, not five.
1 of 3
Quoted from huggingface.co on 2026-09-27: At 144.3M parameters, it turns a state, a question, and possible answers… Quoted from huggingface.co, checked 2026-09-27. “ At 144.3M parameters, it turns a state, a question, and possible answers into one clear decision.” quoted from huggingface.co · checked 2026-09-27

Models · 27 SEP

Supersonic Labs releases Julia-1, a 144.3M-parameter decision model with a 550.5 MiB checkpoint

Built on mmBERT-small, the Apache-2.0-licensed classifier takes a state, a question and up to 20 options, and returns one of them.

Checked 27 Sep, 04:30 · Hugging Face (Supersonic Labs model card) · 1 source
Quoted from code.claude.com on 2026-09-27: Added Claude Opus 5.5 ( claude-opus-5-5 ), now the default Opus model… Quoted from code.claude.com, checked 2026-09-27. “ Added Claude Opus 5.5 ( claude-opus-5-5 ), now the default Opus model — 1M context, $4/$20 per Mtok with $0.20/Mtok cache reads” quoted from code.claude.com · checked 2026-09-27

Tooling · 27 SEP

Claude Code 2.1.280 sets Opus 5.5 as the default Opus model, with a 1M-token window

The changelog lists Opus 5.5 at 4 dollars per million input tokens and 20 dollars per million output tokens.

Checked 27 Sep, 04:43 · Anthropic · 1 source
Quoted from arxiv.org on 2026-09-27: A single production-scale unit of DSec spans around 160 nodes, serving about… Quoted from arxiv.org, checked 2026-09-27. “ A single production-scale unit of DSec spans around 160 nodes, serving about 3 million sandboxes per day; in production, it supports over 380,000 concurrent sandboxes and sustains over 5,000 sandbox creations per second.” quoted from arxiv.org · checked 2026-09-27

Research · 27 SEP

DeepSeek describes DSec, a sandbox platform for agentic training handling 3 million sandboxes a day

The paper details production infrastructure for agentic training, spanning container, microVM and full-VM backends behind one SDK.

Checked 27 Sep, 04:30 · arXiv (DeepSeek) · 1 source
No notice today matches. The archive goes back further. Open the archive
Quoted from huggingface.co on 2026-09-27: At 144.3M parameters, it turns a state, a question, and possible answers… Quoted from huggingface.co, checked 2026-09-27. “ At 144.3M parameters, it turns a state, a question, and possible answers into one clear decision.” quoted from huggingface.co · checked 2026-09-27 Models · 27 SEP Supersonic Labs releases Julia-1, a 144.3M-parameter decision model with a 550.5 MiB checkpoint Hugging Face (Supersonic Labs model card) · 1 source · 27 Sep, 04:30 Quoted from code.claude.com on 2026-09-27: Added Claude Opus 5.5 ( claude-opus-5-5 ), now the default Opus model… Quoted from code.claude.com, checked 2026-09-27. “ Added Claude Opus 5.5 ( claude-opus-5-5 ), now the default Opus model — 1M context, $4/$20 per Mtok with $0.20/Mtok cache reads” quoted from code.claude.com · checked 2026-09-27 Tooling · 27 SEP Claude Code 2.1.280 sets Opus 5.5 as the default Opus model, with a 1M-token window Anthropic · 1 source · 27 Sep, 04:43 Quoted from arxiv.org on 2026-09-27: A single production-scale unit of DSec spans around 160 nodes, serving about… Quoted from arxiv.org, checked 2026-09-27. “ A single production-scale unit of DSec spans around 160 nodes, serving about 3 million sandboxes per day; in production, it supports over 380,000 concurrent sandboxes and sustains over 5,000 sandbox creations per second.” quoted from arxiv.org · checked 2026-09-27 Research · 27 SEP DeepSeek describes DSec, a sandbox platform for agentic training handling 3 million sandboxes a day arXiv (DeepSeek) · 1 source · 27 Sep, 04:30

That is the day: 3 notices.Archive

Today's notices 3 NOTICES

Today's notices

Sun 27 Sep · 3 NOTICES
  1. Models · 27 SEP

    Supersonic Labs releases Julia-1, a 144.3M-parameter decision model with a 550.5 MiB checkpoint

    Quoted from huggingface.co on 2026-09-27: At 144.3M parameters, it turns a state, a question, and possible answers… Quoted from huggingface.co, checked 2026-09-27. “ At 144.3M parameters, it turns a state, a question, and possible answers into one clear decision.” quoted from huggingface.co · checked 2026-09-27

    Built on mmBERT-small, the Apache-2.0-licensed classifier takes a state, a question and up to 20 options, and returns one of them.

    Checked 27 Sep, 04:30 · Hugging Face (Supersonic Labs model card) · 1 sourceRead

  2. Tooling · 27 SEP

    Claude Code 2.1.280 sets Opus 5.5 as the default Opus model, with a 1M-token window

    Quoted from code.claude.com on 2026-09-27: Added Claude Opus 5.5 ( claude-opus-5-5 ), now the default Opus model… Quoted from code.claude.com, checked 2026-09-27. “ Added Claude Opus 5.5 ( claude-opus-5-5 ), now the default Opus model — 1M context, $4/$20 per Mtok with $0.20/Mtok cache reads” quoted from code.claude.com · checked 2026-09-27

    The changelog lists Opus 5.5 at 4 dollars per million input tokens and 20 dollars per million output tokens.

    Checked 27 Sep, 04:43 · Anthropic · 1 sourceRead

  3. Research · 27 SEP

    DeepSeek describes DSec, a sandbox platform for agentic training handling 3 million sandboxes a day

    Quoted from arxiv.org on 2026-09-27: A single production-scale unit of DSec spans around 160 nodes, serving about… Quoted from arxiv.org, checked 2026-09-27. “ A single production-scale unit of DSec spans around 160 nodes, serving about 3 million sandboxes per day; in production, it supports over 380,000 concurrent sandboxes and sustains over 5,000 sandbox creations per second.” quoted from arxiv.org · checked 2026-09-27

    The paper details production infrastructure for agentic training, spanning container, microVM and full-VM backends behind one SDK.

    Checked 27 Sep, 04:30 · arXiv (DeepSeek) · 1 sourceRead

Models 27 SEP

Supersonic Labs releases Julia-1, a 144.3M-parameter decision model with a 550.5 MiB checkpoint

Built on mmBERT-small, the Apache-2.0-licensed classifier takes a state, a question and up to 20 options, and returns one of them.

Hugging Face (Supersonic Labs model card) · 1 source·checked 27 Sep, 04:30
Apache licenceJuliaSupersonic LabsmmBERT
Quoted from huggingface.co on 2026-09-27: At 144.3M parameters, it turns a state, a question, and possible answers… Quoted from huggingface.co, checked 2026-09-27. “ At 144.3M parameters, it turns a state, a question, and possible answers into one clear decision.” quoted from huggingface.co · checked 2026-09-27
Tooling

Claude Code 2.1.280 sets Opus 5.5 as the default Opus model, with a 1M-token window

The changelog lists Opus 5.5 at 4 dollars per million input tokens and 20 dollars per million output tokens.

27 SEP · Anthropic · 1 source
Research

DeepSeek describes DSec, a sandbox platform for agentic training handling 3 million sandboxes a day

The paper details production infrastructure for agentic training, spanning container, microVM and full-VM backends behind one SDK.

27 SEP · arXiv (DeepSeek) · 1 source
The record 1 MORE TODAY

Short of the target: three today, not five.

read 00 of three
drag to begin drag left or right · tap to read more hover to raise · tap to bring one forward tap again to return it to the fan tap a card to expand it tap again to collapse tap a headline to open it in the deck