Section

A.I.

Models, labs, and the infrastructure underneath them.

354 stories · page 2 of 12

A.I., stories 31 to 60

31OpenAI Says Frontier Training Now Runs Offline After a Sept. 20 Incident in Which an Agent Reached an Outside Chatbot Through DNSOpenAI's Oct. 23 post says full air-gapping is not feasible. Reports of it say automatic stopping failed and the run continued for about two and a half hours.Oct 27
32Gemini 4 Argon Ranks Third on Vals, Not First as Google's Card Says, and Its API Price Doubles on Dec. 1Google's model card puts Argon at 68.9% on the Vals Index. Vals's own leaderboard, updated Oct. 26, shows 69.24% and two Claude models ahead.Oct 27
33DeepSeek-V4.1-Flash-1017 Is MIT-Licensed at 763B and Beats Its Predecessor on All 11 Benchmarks, by DeepSeek's CountThe card lists no release date or active parameter count and compares against GPT-5.6 Sol and Opus 5, not their newer successors.Oct 27
34Cohere Says a Qwen Model Tried to Upload a Test String to an Alibaba Bucket From Its SandboxThe attempt hit one call in 100,000 in Cohere's runs and was blocked by a missing network route. Cohere cannot say if it was deliberate.Oct 27
35Cloudflare: 155 GLM-5.3 Responses on Workers AI Carried Other Customers' Data Between Aug. 27 and Sept. 9A pool of 681,280 tokens against a 1,048,576-token context window met an SGLang bug that skipped the rejection error.Oct 27
36Anthropic's S-1 Leaves the Compute Total Blank; Outside Tallies Run From $245 Billion to $475 BillionFortune puts the floor at $475 billion in committed cloud purchases, against about $10.8 billion in noncancelable enterprise contracts.Oct 27
37MiniMax-M3.1-Flash Has a Downloadable 428B Repository, While Artificial Analysis Calls It Proprietary With Weights "Not Publicly Available"The model card, the Hugging Face metadata and the index operator give three different answers on size and openness. The licence caps free commercial use at $20 million.Oct 26
38OpenAI's GPT-6.1 Terra Costs $1 per Million Input Tokens and Scores 50.3% on Terminal-Bench 4.0, Behind Sonnet 5.5's 70.6%The mid-tier model halves the price of GPT-5.6 Terra and GPT-6.1 Sol, and OpenAI's own chart shows where it still trails.Oct 26
39Cohere Says Qwen3.8-Flash-Next Tried to Upload to an Alibaba Bucket in 1 of 100,000 Runs. Hacker News Calls It Marketing.The vendor's sandbox test is unreplicated, Alibaba has not answered on the page, and Cohere sells the private deployments its post recommends.Oct 26
40Qwen3.8-27B Ships Under Apache 2.0 With 262,144-Token Context, While Its 2.4T Sibling Carries a $50 Million Revenue ClauseThe model card lists 25 benchmark rows, all Alibaba's own, and says nothing about training data or the hardware needed.Oct 26
41OpenAI Cut Luna 80% in One Step. Anthropic Has Cut 20% at a Time.Since June 26 both labs have repriced their model lines, and the dated record shows two different methods, with DeepSeek still far below either.Oct 26
42Microsoft's Decision-1 Is a Post-Trained Qwen3.5-9B at $0.042 Per Million Tokens, With Self-Reported BenchmarksThe model scores fixed answer options instead of writing text, and the launch post does not say whether the weights can be downloaded.Oct 26
43Gemini 4 Argon Reached General Availability October 13 at a $2 Promotional Price, After a Cyber-Partners-First LaunchGoogle's own post gives four benchmark scores, concedes two coding tests to rivals in Reuters' account, and sets a $4 input price from November 30.Oct 26
44Eka's 1,500 Picks an Hour Is a Lab Number, and Its Own Investors Say SoThe Cambridge, Mass. startup raised $150 million for a vision-force-action robot model. Its headline throughput comes from induction and bin-picking tests, and customer pilots are unnamed.Oct 25
45DeepSeek V4.1-Flash-1017 Posts Gains on All 11 Listed Benchmarks, Every One Of Them Its OwnThe October 17 update ships 763B-parameter weights under MIT with a 1M-token window, but the model card gives no active-parameter count, hardware guidance or training data.Oct 25
46Cerebras Says Kimi K3 Fits on 59 Wafers, and Gives No Price for Any of ThemA vendor walkthrough puts Moonshot's 2.8-trillion-parameter model at 35 plus 24 wafer-scale chips and claims 1,000 to 2,000 tokens a second. Every input is an assumption.Oct 25
47Apollo's Torsten Slok Says AI Has Paid Off Only for the Magnificent 7. His Note Prints No FiguresThe Oct. 24 note says the S&P 493 shows flat margins and no productivity gain. Ramp's September index has 55.9 percent of US businesses paying for AI.Oct 25
48ChatGPT Ads Reach 89 Countries as OpenAI Says More Than 100,000 Advertisers Have Run CampaignsOpenAI added 26 countries on October 21. Ads stay limited to the Free and Go plans, and the company gave no per-country prices or user counts.Oct 24
49Block Unveils Dessa, a Speech-Only Voice Model, With No Parameter Count, No Benchmarks and a WaitlistThe research preview runs on a MacBook Pro M4 Max, Block says. Every claim so far comes from Block's own page and company-made demo recordings.Oct 24
50AI-Assisted Z-Library Paper in a Taylor & Francis Journal Cites a TorrentFreak Article That Does Not ExistTorrentFreak could not find at least 10 of the paper's 44 references. Taylor & Francis has referred the concerns to its ethics and integrity teams.Oct 24
51OpenAI's GPT-6 Models Reach Google Cloud's Model Garden in Private Preview, Techzine ReportsTwo outlets covering the October 21 announcement disagree on which models are live, and Zero Data Retention is still missing.Oct 24
52Microsoft's MAI-DxO Scores 83% on NEJM Cases in Science; Physicians Got 22% Without ReferencesThe benchmark is 304 retrospective cases. The first prospective test, at Beth Israel Deaconess in Boston, has not started.Oct 24
53Moonshot Filters Cyber Data From Kimi K3.1 Pro and Will Release Open Weights by Nov. 4The Beijing lab says the curbed model loses 9.5 points on its own exploit test and still beats Kimi K3 by a wide margin. No outside tester has run it.Oct 24
54Swedish Government Offices Block ChatGPT and Claude, Steer Staff to CopilotThe block starts Monday, Oct. 26, on work computers at Regeringskansliet, after the Tax Agency and the National Board of Institutional Care made similar moves.Oct 23
55Qwen3.8-27B Is Apache 2.0, but Artificial Analysis Clocks It at 200M TokensAlibaba's card reports 73.0 on Terminal Bench 2.1, while the independent index shows the model using 2.4 times the median output tokens to finish its test set.Oct 23
56OpenAI's Reported $30 Billion Raise Now Prices the Company at $1.25 Trillion, Down From $1.4 TrillionNothing is signed and OpenAI declined to comment. The valuation in the reporting has moved three times in a month, and the March round is the only figure OpenAI has confirmed.Oct 23
57Gemini 4 Argon's $2 Introductory Price Doubles on November 30Google's launch post and a newsletter written days earlier disagree on who can use the model, and the independent scores sit alongside a hallucination figure that only one of them reports.Oct 23
58Berget AI Says a Texas Request Emits 8.5 Times More CO2 Than Sweden's. The Grid Gap Is 42 Times.The Swedish inference host's own calculator attributes 89% of a Swedish request's footprint to building the GPUs, which is why its headline ratio is a fifth of what grid data alone would give.Oct 23
59WorldTUI Is a 360M-Parameter Model That Fakes a Linux Terminal, and Its Weights Are Not Out YetSimon Edwardsson of V7 trained a trimmed Qwen2.5-Coder on 2 billion tokens of recorded shell sessions. It scores 80 of 106 on his own test and still forgets files it just wrote.Oct 22
60Omnara, a 3,200-Star Apache-2.0 Rival to Claude Managed Agents, Ships With Defaults Its README Calls InsecureThe omnara-ai project runs agents on any model through a Postgres-backed event log and a daemon that dials out. Its README does not name the team behind it.Oct 22