AI
Frontier and open model releases, price and API changes, papers, benchmarks and safety evaluations.

Cohere Embed 5 and Rerank 5 join OpenRouter on existing keys
One API key can now run a full retrieval pipeline, but the announcement gives no per-token prices.
2 weeks agoNov 03 – Nov 04
OpenAI is acquiring Pierre to widen Codex's tools for coding
OpenAI said the acquisition expands Codex's tools for writing, understanding and reviewing code, but did not disclose the price or closing date.
OpenAI pipeline generates patches for 948 Linux kernel flaws
OpenAI said an autonomous pipeline handled the work from discovery to patch generation, but its description does not say whether the patches have been merged into the Linux kernel.
Only two of five EU AI Act assistants beat generic baselines
Purpose-built development did not guarantee better performance, and contestant integration code and historical results are outside the public release.
OpenAI raises Security Bug Bounty rewards and offers bonuses
OpenAI is revising how it rewards independent security researchers, but its announcement summary gives no amounts.
OpenAI preparing next security sprint with Cathedral Systems
Cathedral Systems is taking part, but OpenAI has not disclosed its role or when the work begins.
3 weeks agoOct 26 – Nov 02
OpenAI's Pando backend cuts median push latency by about 80%
OpenAI published internal operating results but did not say whether or when Pando itself will be generally available.
Anthropic open-sources FlashProve to verify AI model outputs
The system checks which model produced a response while keeping the weights hidden, but Anthropic has not given a schedule for using it in Claude services.
OpenAI appoints Dan Driscoll as national security counsellor
The advisory brief extends to cyber resilience and collaboration with US allies, but OpenAI has not named specific projects.
DeepSeek releases research coding agent under an MIT licence
The research code acts on requests and team rules, and DeepSeek has not disclosed a general release date or sale price.
datasette-llm 0.1a9 now uses the llm library's default model
Calls with no model ID raise an error if the fallback model lacks a required API key or is excluded by configuration.
PlanAssist AI housing tool expands to 15 councils in England
The 58% cut in officer assessment time is a company-published figure, and the company did not disclose how many applications the pilot covered.
Cohere commits more than C$4 billion for Canada through 2031
The pledge is cumulative from 2024, and Cohere has not disclosed the amount spent so far or how the sum is allocated.
OpenAI strengthens violence prevention safeguards in ChatGPT
OpenAI cites expanded safeguards and earlier intervention, but its summary gives no timing or effect figures for individual measures.
Xbox hardware revenue decreases 14% on lower console volumes
Xbox content and services revenue also fell 2%, and Microsoft has not disclosed the percentage decline in console units sold.
AI agents' experiment ideas clear review in 194 of 415 cases
Evidence that agents' own rankings help decide which experiments to run first is still limited.
Credit Saison's ChatGPT Enterprise pilot delivers a 954% ROI
The figure comes from OpenAI's own customer story, and the story's summary gives no calculation period or cost items.
Google gives cyber defenders first access to Gemini 4 Carbon
Gemini 4 Carbon costs $2 per million input tokens and $10 per million output tokens, but Google DeepMind has not given a date for developers, businesses or consumers.
Anthropic names 20 partners for free Claude research credits
Foundations distribute the credits and research institutions use them in-house, but Anthropic has not disclosed the total.
Legora uses OpenAI models, ChatGPT, and Codex for legal work
OpenAI's introduction sets out why Legora adopted its tools but gives no measure of working time saved.
OpenAI says GPT-6 Astra leads Retell's voice agent benchmark
The introduction to OpenAI's case study cites one benchmark as evidence and gives no sample size or evaluation conditions
Owlchemy Labs lists Hyper Visor game for XREAL AURA at $1.99
Google Play's South Korea page shows the game at ₩2,900, and the official announcement does not state its Korean-language support.
Anthropic offers government 40% off Claude in wider GSA deal
The discount extends to the Claude API, but the summary on the company's page gives neither the period it applies nor the base price.
Google DeepMind opens Genie 4 world model to trusted testers
The model generates scenes and sound together in real time, but Google has not disclosed a general release date or pricing.
Runway makes Ads agent generally available on all paid plans
The agent reads ad performance to build new variants, while a separate analysis questions how far direct publishing is available.
Greptile applies OpenAI's GPT-6 Astra to user workflow tests
OpenAI also described Greptile's use of Daybreak and Codex, with no performance figures in the post's title or summary.
CodeRabbit uses OpenAI's GPT-6.1 Sol to probe security risks
OpenAI did not disclose the security agent's detection accuracy or false-positive rate.
Anthropic files S-1 for proposed Nasdaq IPO under symbol ANT
The share count and price range are undetermined, and no shares may be sold before the registration statement becomes effective.
OpenAI charges $1 per million input tokens for GPT-6.1 Terra
The standard API price is 50% lower than GPT-5.6 Terra's, and OpenAI classifies the model's cybersecurity capability as Critical under its own framework.
Oct 2026Oct 12 – Oct 25
Meta researchers release 120-task test for wearable AI aides
The benchmark scores the final state of a virtual world rather than written answers, but the repository notes list no pass rates by model.
Auditable AI text detector reports small drop in performance
The authors say the detector must not be used for decisions about individuals and that its accuracy beyond its training scope is not established.
OpenAI says Habitat expansion makes ChatGPT reads 12× faster
The company-published figure comes from the Habitat storage system, and the post's introduction gives no comparison point or measurement conditions.
OpenAI says Anima helps clinics cut their phone waits by 75%
The headline and introduction of OpenAI's post do not give the period measured or the number of clinics behind the 75% figure.OpenAI and Poland's NASK cooperate on AI safety and security
The partnership aims to strengthen safeguards, but the announcement summary names no specific projects or duration
OpenAI announces Decisions API for low-cost in-app decisions
OpenAI promotes fast, low-cost decisions, but the announcement's introductory text gives no specific price.
Anthropic seeks life scientists for 8–12 week Claude studies
Anthropic covers the cost of running its models and researchers keep their data rights, but the company has not disclosed how many applicants it will select.
Anthropic donates Agent Skills spec to Agentic AI Foundation
The announcement gives no counting method for the company's adoption figures and no schedule for individual roadmap items.
AT&T legal team rethinks in-house request intake with OpenAI
OpenAI's case study says the aim is easier access to legal knowledge, but its introduction names no model and gives no results.
Cohere releases Rerank 5 at $0.05 per million tokens for Pro
A new feature selects relevant passages to reduce language-model tokens, but Cohere did not say whether it costs extra.
OpenAI links researchers and maintainers on open-source bugs
OpenAI presents the effort as coordination among security researchers, maintainers and incident responders as agents uncover critical vulnerabilities.
Southeast Asia's digital economy totals $368 billion in 2026
The report names a roughly threefold data centre expansion as a condition for AI usage growth, and Google's announcement gives no country figures.
llm-d 1.0 strengthens zero-downtime and multi-tenant serving
The 97% cache reuse and cost-reduction figures are company-published and come from tests whose model and hardware configurations IBM did not disclose.
Inception prices Mercury 3 at $0.20 per million input tokens
Inception said total parameters in its diffusion model grew more than 40 times from Mercury 1, and it has not disclosed when the launch discount ends.
MiniMax unveils M3.1 Flash at $0.15 per million input tokens
MiniMax claims a 70% cost cut for agentic tasks but has not disclosed its previous model's prices or when the weights will be released.
EGI-BENCH puts eight enterprise agent benchmarks in one tool
The authors report that no model leads on every capability axis, but the abstract gives no deep-mode scores.
DeepSeek upgrades V4.1 Flash and leaves API prices unchanged
The company also opened its in-house search API but has not disclosed the framework behind the earlier release's Terminal-Bench 4.0 score.
Junior share of technical hiring at Anthropic declines to 9%
The change came alongside wider use of AI agents in internal company data, and the company noted limits on its productivity estimates.
Survey splits Anthropic staff, US adults over Nobel-level AI
Anthropic's survey found that the two groups broadly agreed on what AI can do today and differed on the pace of future progress.
Microsoft says Magnetar is 10-39 times faster than top zkVMs
The general-purpose proof system rests its security only on standard hash functions, but Microsoft has not disclosed a release schedule or terms of use.
Cohere plans to buy AISLE, finder of 500-plus software flaws
Cohere did not disclose the price or closing date, and set out a plan to expand Prague as an AI security hub.
Four Mac Studios run 816B Muse Spark Open with 4-bit weights
The figures are Apple's own measurements, and the company did not report how 4-bit quantisation affected answer quality.
AI illustrationApple's SSR cuts per-turn reasoning latency by more than 90%
The framework replaces freeform reasoning with a choice among preset candidates to reduce inference costs in small models, and Apple has not disclosed any product plans.
Anthropic reports illicit distillation rate down by over 90%
The reduction is a company-reported figure, and the public summary gives no sample or measurement period.
Gemini 4 Argon solves all 12 simulated cyberattack scenarios
The figures are company-reported with no stated outside replication, and the same post describes attempts to tamper with grading during training.
DeepMind reports Gemini 4 Argon at 87% on physical reasoning
The company's blog post mixes physical robots and simulation in its demonstrations and gives no scores for the models it compares against
Mixing English in pre-training matches 2–13× the Arabic data
Mixing in another language improved pre-training for data-poor languages more than adjusting settings did, the Apple paper said.
Anthropic shifts its code repository to Scion in five months
Anthropic said development infrastructure sized for human workloads had become a bottleneck at agent scale.
Microsoft publishes student simulator for training AI tutors
StudentSim is a research system for use before trials with real students, and Microsoft has not said whether tutors trained with it improve learning outcomes.
Apple's Minions pools idle Mac GPUs into a virtual GPU cloud
An Apple research paper turns GPUs that normally sit idle into inference capacity, targeting the cost and wait times of centralised infrastructure.
Apple paper tests LLMs against humans on translation scoring
The paper argues that LLMs need thorough evaluation across language pairs and domains before they replace human evaluation of translations.
Meta offers Muse Wonder agent model on its API and Muse Code
Performance claims rest on company-published figures, and the announcement gives no API pricing or publication date for the full safety report.
462 in totalWritten by Gyeol HanRSS feed