Site updates
What changed on OrigoTech, date by date: new pages, new data, corrections. The figures themselves stay dated and sourced on each page.
Article: Android Studio Rabbit 2 and AI coding agents
A look at Rabbit 2's third-party agents, skills, code review, LeakCanary and XR emulators.
Android Studio Rabbit 2: Codex, Claude and more AI agents join the IDE
Article: GitHub Agent Merge and blocked pull requests
How GitHub Agent Merge handles CI failures and merge conflicts, with setup, billing, and limitations.
GitHub Agent Merge: can Copilot fix your pull requests while you work?
October 29 roundup: Azure and Google Cloud earnings, Clover-35B
New article on Microsoft and Alphabet earnings (Azure +48%, Google Cloud +103%, AI infrastructure spending), the open diffusion model Clover-35B and new OpenCode, Zed and Cline releases. New key figure: Google Cloud growth. Clover-35B-A3B-Preview joins the model comparison.
Microsoft and Google earnings: Azure grows 48%, Google Cloud doubles as AI spending climbs · Model comparison · OpenCode 2.0 · Zed · Cline · Homepage trends
Article: Genie 4 and interactive world models
New analysis of Google's Genie 4, its interactive simulation, creative uses, limitations and access, with an explanatory diagram.
Genie 4: Google imagines interactive worlds that move, sound and speak
Article: OpenAI and ServiceNow connect enterprise workflows
New analysis of the ServiceNow plugin for ChatGPT and planned Codex, security and voice integrations, with their announced availability.
OpenAI and ServiceNow bring ChatGPT closer to enterprise workflows
Article: Copilot Fusion and the cost of coding agents
New analysis of Copilot Fusion, model orchestration, the Artificial Analysis comparison and GitHub AI Credits.
GitHub Copilot Fusion picks models for you: better coding for less?
RSS feed: full articles and improved dates
The RSS feed now offers full articles in compatible readers, preserves existing links and uses publication times when available.
Analysis: Gemini 4 Carbon, first but not yet available
New article on Gemini 4 Carbon, which is not yet available to developers or the general public: scores against Argon, Opus 5.5 and GPT-6, preliminary LMArena ranking, announced pricing and access limited to the Fairwind Program. Carbon also joins the model comparison, marked "not yet available".
Gemini 4 Carbon: Google retakes the lead but keeps it locked away · Model comparison
Correction: MiniMax M3 licence in the model comparison
MiniMax M3's licence does not ban commercial use: it is allowed after notifying MiniMax, with prior written authorisation above $20M in annual revenue. The model comparison row is corrected, with the licence text as its source.
Analysis: the state of open-weight models in October 2026
New article on open-weight models: who leads, the gap with closed models according to Artificial Analysis and LMArena, prices, licences and Hugging Face downloads, with three charts and two sourced tables.
Open-weight models: who leads, and how far behind closed models
October 28 roundup: Google TPUs at Oracle, Meta's $95B, Gemini API CLI
New article on Google and NVIDIA chips in Oracle's cloud, Meta's financing plan, the Gemini API CLI for agents, new versions of Codex, Claude Code and OpenCode, and Liquid AI's LFM3. New key figure: Meta's $95B plan. LFM3-1.7B joins the model comparison.
Oracle opens its cloud to Google TPUs, Meta plans $95B of stock for AI: the October 28 roundup · Model comparison · Claude Code · Codex · OpenCode 2.0 · Homepage trends
Analysis: Anthropic's economic strategy ahead of its IPO
New analysis of Anthropic's S-1: revenue, a mostly accounting loss, compute commitments and governance, with three sourced charts.
October 27 roundup: Anthropic's S-1, GPT-6.1 Terra, chip financing
New article on Anthropic's IPO filing, GPT-6.1 Terra, GitHub Copilot's assisted permissions and the chip financing deals for Anthropic (Broadcom) and SpaceX. New key figure: Anthropic's $16B of revenue in the first half of 2026. GPT-6.1 Terra joins the model comparison.
Anthropic files to go public, OpenAI launches GPT-6.1 Terra: the October 27 roundup · Model comparison · GitHub Copilot · Homepage trends
Self-promotion: Rekha, the author's game
New Rekha page, an iPhone game by OrigoTech's author where every level is an Eulerian path to trace in a single gesture. A box labelled “Self-promotion” appears at the end of articles and at the bottom of the homepage, with a link in the footer; no third-party script, cookie or tracker is added.
October 26 roundup: single agent, Nanbeige4.3-33B, Cline Desktop
New article on Anthropic's guide recommending folding subagents into a single agent, Nanbeige4.3-33B, Cline Desktop 0.0.49, delay penalties at AI data centers and Higgsfield's raise. New key figure: more than $5M in penalties per day of delay. Nanbeige4.3-33B joins the model comparison.
Anthropic makes the case for a single agent, Nanbeige4.3-33B, Cline Desktop 0.0.49: the October 26 roundup · Model comparison · Homepage trends
This week's figures: GitHub repos and prices rechecked
Stars, forks and last-push dates of the 20 repositories updated via the GitHub API (OpenCode reaches 221,647 stars; OpenHands moves back ahead of the MCP reference servers). Model prices, context windows and licences in the comparator, and the plans on tool pages, were rechecked on the official pages on October 26, 2026: no price changed. The note on GitHub Copilot's Free plan is corrected.
October 25 roundup and a homepage focused on the latest
New article on OpenAI's discounts against Anthropic (The Information investigation), Block's Dessa, AWS's Infinity runtime, Ollama 0.40.11 and Qwen Code 0.26.1. New key figure: 100 KB of memory per idle agent on Infinity. The homepage now features the most recent story, lists the 9 latest news articles, shows the most recently verified figures and the latest directory updates; guides and older market benchmarks move further down. New "All articles" page.
OpenAI wins Anthropic customers on price, Block's Dessa, AWS Infinity, Ollama speaks: the October 25 roundup · All articles · Homepage trends
Roundup: Cursor Computer Use, Claude Enterprise, OpenAI offline, Agensh
New article on Computer Use in Cursor, Claude Enterprise's unified chat and Cowork experience, OpenAI's Decisions API ($0.10 per million input tokens), OpenAI's frontier training with no internet access and Microsoft's Agensh (up to 1,024 agents). New key figure: Firebird's estimated $5 billion investment in Kazakhstan. Latest version and sources updated on the Claude Code (v2.1.310), Zed (v1.25.3) and Cursor pages.
Cursor Computer Use, unified Claude Enterprise, OpenAI trains offline, Agensh: the October 25 roundup · Cursor · Claude Code · Zed
Roundup: Kimi K3.1 Pro, Codex CLI 0.164, Copilot in Teams, Replit
New article on Moonshot AI's Kimi K3.1 Pro (1 million tokens, weights promised by November 4), Codex CLI 0.164.0, GitHub Copilot in Microsoft Teams and Replit's October 23 updates. New key figure: Cambricon's +147.7% net profit in Q3 2026. Kimi K3.1 Pro joins the model comparison; latest version and sources added to the Codex, GitHub Copilot and Replit Agent pages.
Kimi K3.1 Pro, Codex CLI 0.164, Copilot in Teams, Replit: the October 24 roundup · Model comparison · Codex · GitHub Copilot · Replit Agent
Roundup: Flint, Replit Monitoring Agent, EGM plugin, Claude on Google Cloud
New article on Flint (Recursive's local coding model, Apache 2.0), Replit's Monitoring Agent, Google's open-source Elephant-Goldfish Model plugin and the Anthropic-managed Claude option on Google Cloud. New key figure: Nokia's +123% sales to AI and cloud customers in Q3 2026. Flint joins the model comparison; sources added to the Replit Agent and Claude pages.
Flint runs locally, Replit Monitoring Agent, Google's EGM plugin, Claude on Google Cloud: the October 23 roundup · Model comparison · Replit Agent · Claude
The OrigoTech newsletter gets the new design
Newsletter emails now follow the site's identity: IBM Plex Sans and JetBrains Mono, an orange accent, thin rules and figures set in monospace, with a readable dark mode. On mobile, the signup form now shows the field and button at full width, and an invalid address is flagged more clearly.
New visual identity, more editorial and more technical
The site adopts a tech-publication identity: IBM Plex Sans for text and JetBrains Mono for data (dates, prices, figures, sources), fonts served from the site itself, a black-and-white palette with a single orange accent, thin rules instead of shadows, one lead story followed by a dense article list, and more readable tables. Content and figures are unchanged; on mobile, the header no longer overflows at 375 px.
Model comparison · Price vs performance of coding models · Stack Overflow 2026: agents in, production off-limits
Article: Agent Skills joins the Agentic AI Foundation
New article on the donation of Agent Skills, Anthropic's SKILL.md format adopted by more than 70 agent platforms, to the Agentic AI Foundation (Linux Foundation), which also takes on Open Responses, OpenEnv and OpenSandbox and now hosts 10 projects. Related article added to the Claude Code, Codex, GitHub Copilot and Cursor pages.
Agent Skills moves to the Linux Foundation: Anthropic hands its standard to the AAIF · Claude Code · Codex · GitHub Copilot · Cursor
Roundup: VS Code 1.143, GPT-6 on Google Cloud, NVIDIA DSX POD
New article on VS Code 1.143 (model comparison, on-device model), GPT-6 arriving in Google Cloud's Model Garden and NVIDIA's DSX POD. New key figure: the 14-day deployment NVIDIA announces at a prepared site. Sources added to the GitHub Copilot, Codex and GPT pages.
VS Code 1.143, GPT-6 on Google Cloud, NVIDIA DSX POD: the October 22 roundup · GitHub Copilot · Codex · GPT
Three articles: DeepSeek and Huawei, AI data centers, Gemini 4 Argon
New articles on the reported DeepSeek-Huawei co-design of Ascend chips (anonymous sources), on the paper published by Brookings that puts the revenue needed to pay off AI data centers at $3.725 trillion a year by 2032, and on the 25 trillion tokens Gemini 4 Argon processed in 72 hours. Related articles added to the Gemini 4 Argon and DeepSeek V4.1-Flash-1017 pages.
DeepSeek is reportedly helping Huawei design its next AI chips: what is sourced and what is not · AI data centers: $3.7 trillion a year in revenue by 2032 to pay off, according to a paper published by Brookings · Gemini 4 Argon: nearly 25 trillion tokens in 72 hours, three times Gemini 3 · Gemini 4 Argon · DeepSeek V4.1-Flash-1017
News: open-source tooling for agents and inference
New article on Microsoft Scope, Vercel durable-harness, DeepSeek DeepInfer and vLLM Humming. New key figure: $38 revenue per GPU-hour announced by DeepSeek for V4.1 Flash; DeepInfer source added to its tool page.
Scope, DeepInfer, Humming: tooling for agents and inference goes open source · DeepSeek V4.1-Flash-1017 · Key figures
Model comparison: three additions and firmer data
Amazon Nova 2 Lite, Granite 4.2 30B (IBM) and Phi-4-reasoning-vision-15B (Microsoft) join the model comparison, now 33 models. GPT-6.1 Sol release date added; MiniMax M3.1 Flash weights confirmed unpublished, with source.
Benchmarks and Updates pages
New page grouping by benchmark every coding score already in our data (11 vendor-reported DeepSWE v1.1 scores), with conditions and sources, and this site changelog, also included in the RSS feed.
Five practical guides
New step-by-step guides: installing Claude Code, setting up an MCP server, running a coding model locally with Ollama, choosing an agent for your budget, moving from Gemini CLI to Antigravity CLI. Linked from the relevant tool pages.
How to install and get started with Claude Code: a step-by-step guide · How to set up an MCP server in Claude Code, Cursor and VS Code · How to run a coding model locally with Ollama · Choosing a coding agent by budget: free, $20 a month or enterprise · Moving from Gemini CLI to Antigravity CLI: installation and migration
Price / performance chart and cost calculator
Two interactive pages attached to the model comparison: a scatter plot of blended price against DeepSWE v1.1 score (10 models) and a monthly cost calculator based on tasks and tokens. No new figures: everything comes from the comparison data.
Price vs performance of coding models · AI model API cost calculator
“X vs Y” and “Alternatives to X” pages
51 head-to-head pages and 23 alternatives pages generated from the sourced tool pages, in French and English, linked from the tool pages and the Comparisons page.
Model comparison: 30 models
24 models added to the comparison (30 rows in total) with price, context, license, release date and vendor-published DeepSWE v1.1 score, sources verified on October 20, 2026. New vendor filter.
“GitHub repos” section
New page listing 20 AI GitHub repositories for developers (stars, forks, license, last push, GitHub API figures verified on October 20, 2026) and two related articles.
AI GitHub repositories · The most-starred AI GitHub repositories among developers (October 2026) · Open-source coding agents on GitHub: OpenCode, Cline, OpenHands, Aider
Junie, Replit Agent, Lovable and v0 pages
Four new tool pages with pricing verified on October 20, 2026, and a new “App builders” category in the directory.
Junie · Replit Agent · Lovable · v0
October audit: comparison, glossary, trust
First model comparison, four written comparisons, Aider, Cline, Zed and CodeRabbit tool pages, Glossary, Contact and Corrections policy pages, per-page share images and richer structured data.
Tool page sources
Every directory page now lists its official sources, all verified on October 20, 2026.
In-depth articles
Seven articles expanded (500 to 700 words), with every figure linking to its re-checked primary source.
Gartner's first Magic Quadrant for coding agents · Stack Overflow 2026: agents in, production off-limits
Author page
Author page with a short bio and the list of signed articles; article bylines now link to it.
Tool page details
Context window, MCP support, operating systems and languages filled in on tool pages where official documentation states them.
New fields and links on tool pages
Tool pages gain context, MCP, operating system and language rows, plus a section of related articles and comparisons.
More visible RSS feed
RSS feed link in the footer and the usual addresses (/feed, /rss…) redirected to it; tags used by a single article are no longer indexed.
Real dates in the sitemap
The sitemap now gives each page's real last-modified date; tool pages accept a list of sources.
Trend sources
Key figures and trend charts now link to their primary source, with a verification date.
Update dates
Update date shown on re-checked tool pages and articles; Claude Code platforms corrected.
Newsletter, methodology, sourced figures
Newsletter sign-up, Methodology page, dated and sourced figures, accessible trend tables.
Week 43 analysis
First weekly long-form analysis: OpenAI, COBOL and low-resource languages.
Excerpts and corrections
Article summaries brought to 120–160 characters, Veracode citation and Gartner attributions corrected.
OrigoTech v2
Search, directory filters and detailed tool pages, comparisons, RSS feed, topics and tags, related articles.
Search · Comparisons · Topics
Security, legal pages and accessibility
Security headers, legal notice and privacy policy, accessibility and SEO fixes.
October 20 articles and five tool pages
Two articles (agents week, Flash model prices), five tool pages (OpenCode 2.0, Muse Spark Open, MiniMax M3.1 Flash, DeepSeek V4.1-Flash-1017, Gemini 4 Argon) and a corrected Veracode source.
Week of October 12: coding agents go team-mode · MiniMax and DeepSeek: Flash models slash the price of agentic coding
OrigoTech goes live
Launch of the bilingual site on AI for code: five articles, a first tool directory and key trend figures.