Models
CONFIRMED
3 min
OpenAI API page prices three GPT-6 tiers at the same context ceiling
The OpenAI API page prices GPT-6 Astra at $10.00 per 1M input tokens and $50.00 per 1M output tokens, with 1.05M context.
Models
Lab releases, prices, and the page the number came from.
21 briefs on the desk
Models
CONFIRMED
1 min
Microsoft-Decision-1's Qwen3.5-9B base is a stopgap
Microsoft says the Qwen3.5-9B version of Decision-1 is a stopgap, because it will rebase the model on its own and OpenAI models.
Models
SINGLE-SOURCE
2 min
Proteus reports over 99 percent phishing recall on a one-week AegisAI sample
AegisAI's Proteus identified more than 99% of 3,020 held-out phishing attacks, while a closed-source model missed 565 and an open-weight model missed 604.
Models
MULTI-SOURCE
1 min
TypeSafe raises $870 million at a $7.5 billion valuation
TypeSafe raised $870 million at a $7.5 billion valuation in a round led by Andreessen Horowitz, with Sequoia Capital and DCVC participating.
Models
CONFIRMED
1 min
A false homicide tip from an Anthropic model was marked as spam
The Philadelphia Police Department says the tip, submitted through PhillyUnsolvedMurders.com, was flagged as spam and never forwarded to the Real-Time Crime Center.
Models
CONFIRMED
2 min
Anthropic's free OSS scanner leaves model metrics unmeasured
Anthropic expects the free OSS scanner to reach above 90 percent accuracy, but reports ship without human review for opted-in open-source projects.
Models
SINGLE-SOURCE
1 min
OpenAI revenue gap is an accounting question, not a demand collapse
The market is pricing an accounting gap: OpenAI's roughly $50 billion annualized revenue rate at the end of September sits below the earlier $70 billion figure, while the company negotiates at least $30 billion in new capital.
Models
SINGLE-SOURCE
2 min
IQuest-Q1's 320B MoE leaves self-hosting unpriced
IQuest-Q1 is a public 320B-parameter sparse MoE model with about 15B active per token, but the release leaves context length, latency, and hardware vendor unmeasured.
Models
SINGLE-SOURCE
2 min
Falcon-ASR's Emirati focus leads the Arabic WER story
Emirati focus leads the Falcon-ASR story, while the Hugging Face filing gives a 20.92 percent average word error rate across six Arabic test sets.
Models
CONFIRMED
2 min
EmbeddingGemma 2 is a label, not a measurement
The vendor page names EmbeddingGemma 2 as an open, lightweight multimodal embedding model, but it gives no benchmark, parameter count, license, or release date.
Models
SINGLE-SOURCE
2 min
Healthleap's hospital risk score is a workflow bet
Hospital records, not a named model, are the bet behind Healthleap's round.
Models
CONFIRMED
1 min
OpenAI memo says terminations were not about raising safety concerns
An internal memo, carried by TechCrunch, says the terminations were not about raising safety concerns, while three OpenAI safety researchers published a letter disputing misconduct claims.
Models
CONFIRMED
3 min
Nano Banana 2 Lite and Gemini Omni Flash named without figures
The DeepMind blog page names Nano Banana 2 Lite and Gemini Omni Flash and provides no parameters, no context window, no price, and no benchmark.
Models
SINGLE-SOURCE
2 min
USA Today filing asks OpenAI for more than $250 million
USA Today Co. filed a copyright suit against OpenAI seeking more than $250 million, alleging hundreds of thousands of articles were copied for model training.
Models
SINGLE-SOURCE
2 min
Biohub's $1.8 billion virtual biology commitment
The release puts $1.8 billion into the Virtual Biology Initiative, with DOE adding more than $500 million over five years.
Models
CONFIRMED
2 min
ChatGPT for Teens leaves the parent alert path unverified
Parent alerts are the missing piece in ChatGPT for Teens, not the model's ability to recognize a crisis prompt.
Models
CONFIRMED
2 min
Anthropic lists a 2026 usage policy update without a figure
A 2026 Usage Policy update appears in the Anthropic newsroom on Oct 8, 2026, beside the Cyber Mission and scientific discovery announcements.
Models
SINGLE-SOURCE
2 min
Meta's signposting tools now look at ad destinations
Where an ad sends a user is what Meta's new signposting tools now examine, as the company says it is trying to catch ads that may point people toward child sexual exploitation content outside its platforms.
Models
CONFIRMED
2 min
Exclusive access, not proof, frames Biohub's $1.8 billion cell-behavior AI push
One year of exclusive data access is the part of Biohub's $1.8 billion cell-behavior AI push that separates commercial funders from public work.
Models
SINGLE-SOURCE
1 min
Melius pivots to creative agents after scrapping first product
After scrapping its first ad-optimization build, Melius has raised $25 million, including a $20 million Series A led by CRV.
Models
CONFIRMED
4 min
GPT-6 ships interactive UI to 1.2 billion ChatGPT users
OpenAI's October 7 release puts GPT-6 with Intelligent UI into ChatGPT's Chat tab, replacing plain text with interactive components for 1.2 billion weekly users.