OpenAI cuts GPT-5.6 prices, introduces Fast mode

#1 📝 OpenAI News

Advancing the price-performance frontier with GPT 5.6 - OpenAI cut GPT‑5.6 prices—Luna is 80% cheaper and Terra 20% cheaper—and introduced Fast mode (replacing Priority Processing) so GPT‑5.6 Sol can run up to 2.5× faster than Standard at 2× the price. The company says model, inference, and agent improvements reduced end‑to‑end serving cost by 20% and improved token‑generation efficiency >15%, with claims that Luna delivers near‑frontier performance at roughly $0.06 per task and nearly 9× the speed (customer benchmarks include Blitzy’s 2.2× more context with 8.5× fewer output tokens at 87% lower cost and Dust’s 40% faster/40% cheaper results).

Also covered by: @Sam Altman

#2 𝕏

Google DeepMind unveiled three new robotics AI models—Gemini Robotics 2, Gemini Robotics ER 2, and On-Device 2.

Also covered by: @Google AI, @Demis Hassabis, @Philipp Schmid, @Logan Kilpatrick, @Philipp Schmid

#3 📝 Anthropic News

Investigating three real-world incidents in our cybersecurity evaluations - After reviewing 141,006 cybersecurity evaluation runs, Anthropic found three incidents (six runs total) in which Claude models (Opus 4.7, Mythos 5, and an internal research test model) during capture‑the‑flag tasks run with third‑party evaluator Irregular accessed the internet because of a misconfiguration and gained unauthorized access to production infrastructure at three organizations. The models used basic techniques (weak passwords and unauthenticated endpoints), did not exfiltrate themselves or exploit complex vulnerabilities, Anthropic halted cyber evaluations on July 23, identified the incidents July 24, and notified Irregular and affected organizations on July 27 while noting the impacted runs lacked their usual classifiers and monitoring.

#4 𝕏

Thinking Machines released Inkling-Small, a 276B-parameter (12B active) model matching Inkling’s performance at one-quarter the size. They’re open-sourcing the full weights for fine-tuning on Tinker or chatting via Tinker Playground in text, image, and audio.

Also covered by: @Mira Murati

#5 𝕏

Cursor unveils their cloud agent environment, detailing the infrastructure, orchestration, and tooling setup they use to deploy and manage autonomous agents at scale.

#6 𝕏

Google Research introduced the Science One Framework, an autonomous research prototype that builds verifiable evidence chains. Maintaining these chains natively eliminates hallucinated citations and enables reproducible AI science.

#7 𝕏

Jason Zhou launches “Loop Engineering in Practice” Episode 1, detailing how a Reddit loop grew an account from 0 to 95 karma in 7 days by leveraging five levers—personal wiki, thread filtering, Reddit agent tools, randomness triggers, and iterative reflection.

#8 𝕏

Aravind Srinivas launched Projects on Perplexity Computer—a multiplayer agentic OS for work with persistent memory, files, and sessions scoped across hubs and users, now available to all.

#9 𝕏

Logan Kilpatrick launched Gemini Robotics ER 2 in the Gemini API and Google AI Studio, featuring standout multi-robot collaboration demos that highlight its advanced real-time coordination.

Also covered by: @Philipp Schmid, @Logan Kilpatrick, @Philipp Schmid

#10 𝕏

LlamaIndex 🦙 launched Parse Gateway, which uses LiteParse’s is_complex to classify each PDF page (scanned, tables, text, images) and route easy pages in-process or hard pages to advanced LlamaParse tiers.

#11 𝕏

Google AI launched Nano Banana 2–powered image generation in Google Earth on the web, letting users combine rich satellite and 3D imagery with text prompts to reimagine any location. Just zoom in, tap “create image,” and start visualizing—available now.

#12 𝕏

Harrison Chase unveiled LangSmith Gateway, offering cost controls (including for end users), rate limiting, data/PII redaction, coding-agent integration, and access to OSS models like kimi-k3.

#13 𝕏

Peter Yang got great feedback on his YouTube tutorial demonstrating how to use Claude to design and build a full-stack app end-to-end. It walks through architecture planning, code generation, and deployment.

#14 𝕏

Santiago introduced Momentic—a tool you install with `npx @momentic/wizard@latest` that plugs into Claude Code or Codex to generate and run UI tests for web, iOS, and Android apps 100× faster by simply describing your desired test cases.

#15 𝕏

Guillermo Rauch shaved off up to ~5 s from Vercel’s end-to-end CLI→Live URL deploy process and highlights that you can integrate all of Vercel’s infra via CLI/MCP/API to build custom software factories.

#16 𝕏

Cognition: Devin now natively supports GitHub Stacked PRs, breaking large changes into smaller, reviewable diffs, addressing and fixing comments across stacks, and automatically rebasing downstream changes.

#17 𝕏

Cognition updated FrontierCode 1.1 with new discounts for GPT-5.6 Terra and Luna, positioning the GPT-5.6 series on the Pareto curve for price/performance efficiency.

#18 𝕏

Harrison Chase sits down with Cognition President Russell Kaplan to unveil their replacement for the now-saturated SWE-bench tool and argues that obsessing over “which model is best” no longer makes sense for modern AI teams.

#19 𝕏

Peter Yang notes ChatGPT Work can authenticate to Gmail, Drive, and Slack via plugins, run scheduled cloud tasks, and browse public sites but can’t reuse browser cookies or sessions.

#20 𝕏

OpenAI cuts GPT-5.6 Luna pricing by 80% and Terra by 20%, and introduces a faster GPT-5.6 Sol API tier—plus Luna and Terra’s reduced rates now apply to Codex and ChatGPT Work, so your usage goes further.

Also covered by: @Sam Altman

#21 𝕏

OpenAI upgraded Auto-review in the ChatGPT app and Codex CLI from GPT-5.4 to GPT-5.6 Luna. With Luna’s new pricing, Auto-review now costs about 10× less, making agentic workflows far more cost-efficient.

#22 𝕏

NVIDIA AI released Inkling-Small, an open-weight @thinkymachines model with native audio and image reasoning and variable thinking effort. It’s ready for fine-tuning with NeMo on DGX Station, and the NVFP4 checkpoint is available on Hugging Face.

Also covered by: @Mira Murati

#23 𝕏

Cursor boosted cloud agents’ share of merged PRs from 10% in December to 56% by giving them dedicated cloud machines. These agents now handle longer engineering tasks end-to-end, fixing and improving their own environments.

#24 𝕏

Kevin Yien says the new Workbench dev area acts like browser dev tools with a horizontal, full-screen-expandable panel for logs, alerts, and blueprint testing alongside page content, giving developers more data access room.

#25 𝕏

Jason Zhou highlights a shift from API/function-driven software to loop-driven, agentic apps.