Noticiário do Joãozinho: Edition no. 11 · Nov. 5, 2026 · Morning

High importanceNov. 4, 2026

Cognition launches SWE-2 Flash (276B-A12B): 1/5 the price and 3× the speed of SWE-2, weights still not public

Cognition unveiled (11/04) SWE-2 Flash, a US coding model at 276B-A12B. For the first time the recipe includes mid-training before RL, starting from Inkling-Small (Thinking Machines): Terminal-Bench 4.0 rises from 0.9% to 20.0% and FrontierCode 2.0 from 3.2% to 24.0%. On leading evals, it beats US open-weight models even ~3× larger. On Devin Desktop and CLI, it costs 1/5 and runs at 3× the speed of SWE-2 (about 200 vs 60 tok/s). ExploitBench rises from 11.3% to 34.8%. Weights are still not public; Cognition is building a secure sharing program with partners and researchers.

High importanceNov. 4, 2026

OpenAI validates 948 Linux LPEs with an autonomous cyber pipeline and “upcoming cyber model”; API cost of US$1.213 million

In the post “A thousand fewer paths to root” (11/04), OpenAI describes an autonomous pipeline for discovering, exploiting and patching vulnerabilities, powered by an “upcoming cyber model” (still without a public product name). In a mid-October 2026 run, the system flagged 2,308 candidates, dynamically reproduced 1,724 vulnerabilities and produced 948 local privilege escalation (LPE) exploits on Linux, plus 291 information-leak PoCs. Total run cost was US$1,213,430 in API credits — about US$1,279 per LPE. On an adapted kernelCTF environment, it got 21 LPEs and 25 info-leak PoCs for about US$32,936. The team says it sampled about 10 new exploitation techniques. Patches go through Patch the Planet, with human review before reporting to maintainers.

High importanceNov. 4, 2026

OpenAI announces acquisition of Pierre Computer Company, maker of Code Storage and Codex’s Trees and Diffs libs

OpenAI announced (11/04) it will acquire Pierre Computer Company, Inc. (“Pierre”), behind Code Storage and open-source developer tools. Trees and Diffs — used to browse files and review changes in the Codex app — already come from Pierre. Financial terms were not disclosed; closing depends on regulatory approvals and customary conditions. Until then, the companies remain separate. Pierre’s open-source projects stay open-source, and Code Storage customers can keep using the service while OpenAI brings the capabilities into Codex. The announcement cites OpenAI’s recent work on Pando, an internal source-control backend for engineers and agents.

High importanceNov. 4, 2026

Conductor joins Meta; Cloud shuts down on 11/20 and the local app continues into early 2027

Charlie from Conductor announced (11/04) that the multi-agent coding-agents startup is joining Meta. No deal value was disclosed. From 11/04 there are no new sign-ups; existing free users cannot upgrade to pro, and current pro users keep the features for free. On 11/13 Conductor Cloud goes read-only; on 11/20 Cloud shuts down and data deletion begins — the company asks users to export first. Local Conductor keeps getting new models and critical fixes into early 2027; Meta/Conductor promises a new product early next year. Active subscribers get a refund.

High importanceNov. 4, 2026

Ricursive releases Ripple-65B Preview: dense 65B open-weight, US SOTA on agentic coding (TerminalBench 2.1: 80.52)

Ricursive Intelligence launched Ripple-65B Preview, an open-weight variant of the model that powers its chip-design platform. Dense (not MoE), 65B, 80 layers, d_model 5120, pre-trained on 12 trillion tokens with up to 1M context. It marks US open-weight SOTA on agentic coding on TerminalBench 2.1 at 80.52 — ahead of Reflection Beam 501B (80.1), Meta Muse Spark Open 816B (76.4) and other larger models. Total R&D cost (from scratch): US$32 million. Available on Hugging Face, Together AI, GMI Cloud and OpenRouter; Fireworks the same day. Runs on DGX Spark-class hardware.

High importanceNov. 4, 2026

HumanLayer joins Factory; product will go open-source and multiplayer tasks/plans move into the app

Factory announced (11/04) that HumanLayer, from Dex Horthy and Kyle Mistele, is joining the company. Terms were not disclosed. Factory says it will open-source HumanLayer and invite users to Factory; during the transition, support covers exporting data, migrating to Factory and/or self-hosting the open-source. In the coming weeks, the team plans to bring into Factory features such as multiplayer tasks/workstreams and in-app collaboration on plans, product specs and design docs generated by agents.

XHigh importanceNov. 4, 2026

Rumor/leak: GPT-6.2 Astra and GPT-6.2 Sol show up on the Google Cloud Console quotas page

A post from @testingcatalog shows screenshots where GPT-6.2 Astra and GPT-6.2 Sol appear in the Google Cloud Console quotas UI. Treat as a UI rumor/leak — there is no official OpenAI announcement in this edition. A related post from @lyraxana (https://x.com/lyraxana/status/2118123296738677133) also echoed the theme (~174,195 views, 2,200 likes in the same window).

Engagement 48.6K views·1.1K likes·53 reposts·14 quotes·43 replies·90 bookmarks (as of Nov. 5, 2026 08:07)