everyone's doing the math on terra at $0.96 a task. i did a different math: how many times i'd have to rerun it to get what opus 5.5 gives me first try. cheap per task isn't cheap per result
2010: "touch screen laptops are ergonomically terrible"
2026: "we brought touch to the mac!"
the arm didn't change. the screen still wobbles on the hinge. only the narrative did
5 lines for your CLAUDE.md on a laravel + mariadb project:
1. exact php + mariadb versions
2. "never run migrate:fresh" in caps
3. the one command that runs tests
4. where business logic lives (not controllers)
5. what "done" means: tests green, pint clean
the agent can't guess any of this
60% of the benchmark went private and the open-weights gap quadrupled overnight. turns out a lot of "open model ties the frontier" was the public test set talking
Unpopular opinion: everyone is arguing about vector databases while 90% of real client projects still need one boring MariaDB table with a good index.
artificial analysis rebuilt its index and 60% of it is now private. what moved: open weights went from 3 points behind to 12, and cost per task now spans 13x in the top 10 (terra $0.96, fable 5.1 $13.04). opus 5.5 still leads at 58. honestly terra is DOA for me, cheap doesn't matter if i'd never pick it
the fine print in index v5 that matters for coding: frontiercode 2.0 is graded "on whether a repository maintainer would merge the agent's change". not "tests pass". merge. first benchmark in the index that grades what a client actually pays for.
Announcing Artificial Analysis Intelligence Index v5, featuring 2 new benchmarks and 3 upgrades. For the first time, the majority of the Index is built on private datasets
The Artificial Analysis Intelligence Index is our composite measure of model capabilities across tasks reflecting real-world usage today. v5 is our biggest upgrade since v4 launched in January. With v5, 60% of the Index is now based on evaluations with private test sets or held-out answers.
Alongside Intelligence Index v5, we’re launching the Artificial Analysis Progress Index v1, which tracks model capability over time on one continuous scale.
New to the Intelligence Index:
➤ FrontierCode 2.0: Cognition's newly-launched private benchmark of real-world coding tasks, graded on whether a repository maintainer would merge the agent's change
➤ Terminal-Bench-Science: 70 agentic scientific research tasks, run in a terminal environment
Upgraded:
➤ HLE-Diamond (replacing Humanity's Last Exam): a refined 1,000-question version of Humanity's Last Exam, split evenly between reasoning and knowledge questions
➤ AA-Omniscience v1.1: An updated dataset with saturated and public set questions removed, new questions added, and ambiguous or erroneous questions revised
➤ CritPt v2: a private, 40-question hard subset of CritPt v2, revised by the CritPt authors and run with a code-execution tool
Staying in the Index: AA-Briefcase, AutomationBench-AA, GDP.pdf and Terminal-Bench 4.0
Removed from the Index: GDPval-AA v2.1, SciCode, and AA-LCR v1.1
Key results:
➤ Claude Opus 5.5 (max) leads at 58, ahead of GPT-6 Astra (max, 57) and GPT-6.1 Sol (max, 55): Anthropic and OpenAI hold the top five models. Gemini 4 Argon (high) is next at 50
➤ The open weights gap widens from 3 points to 12: Kimi K3.1 Pro (weights not yet released) is the top open weights model at 46. Among released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) lead at 32 - 26 points behind the leader and below Claude Haiku 5.5 (max, 36)
➤ Cost per task varies by more than 13x among the leading models: of the 10 highest-scoring models shown, GPT-6.1 Terra (max) is the cheapest at $0.96 per Intelligence Index task and Claude Fable 5.1 (max) the most expensive at $13.04
Freelancing in 2026: clients send me the code their AI wrote and ask me to "just fix a few small bugs". The small bug is the architecture.
Codex vs Claude Code, as a freelancer using both daily: Codex does exactly what you asked. Claude Code does what you meant, plus three things you didn't. I still can't decide which one is the feature.
Me: "do not touch the payment flow"
The agent: "Understood. I refactored the payment flow for clarity."
Hot take: PHP is the best language for AI coding agents. 25 years of boring, stable code in the training data. The agent writes Laravel like a senior and your shiny new framework like an intern on day one.
me: has access to the smartest coding models ever made
also me: 14 unfinished side projects, now with better commit messages
terminal vs desktop app is the new vim vs emacs, except now neither side is the one typing the code teapaw.org/leoleotoobad/status/21…
sam altman finding out what every dev learns in their first week on a legacy codebase teapaw.org/pubity/status/21140416…
the 5 stages of AI usage limits:
1. denial
2. anger
3. bargaining
4. a second subscription
5. acceptance (until the reset)
The best coding model isn’t the one with the highest benchmark score.
It’s the one that understands “do not touch the payment flow.”
Guido retweeted
GIVING AWAY 1 FREE COPY OF #ModernWarfareII Vault Edition!!
RETWEET + FOLLOW (so that I can DM you for details) + REPLY with your platform and I will hook up a lucky winner!
