Software engineer. Mostly AI and dev tools, occasionally right.
Tbilisi, Georgia
Joined May 2021
- Tweets36
- Following51
- Followers9
- Likes22
Pinned Tweet
No, OpenAI didn't solve P vs NP.
The claim is narrower. Its model found an algorithm that refutes the Exponential Time Hypothesis, a 1999 conjecture that 3-SAT needs exponential time. There's a Lean proof, and two of ETH's authors are reviewing it.
openai.com/index/building-sc…
ThinkRail, an open-source AI coding IDE from JetBrains' incubator, is trending on GitHub.
Each task gets its own git worktree, so parallel agents never touch your checkout.
Its editor is Monaco. JetBrains using VS Code's editor wasn't on my bingo card.
github.com/JetBrains/thinkra…
Node 26 became LTS on Oct 28, and Temporal works in it without a flag.
Ask Date for Jan 31 plus one month and it says March 3. Temporal says Feb 28, or throws if you tell it not to guess.
Ran on Node 26.13.0.
nodejs.org/en/blog/release/v…
Martin Fowler on LLM voice: the tell isn't the em-dashes, it's "the load-bearing tapestries."
His better point: he often understands a topic only after wrestling it into words. Let a model write the draft and you skip that step.
martinfowler.com/bliki/LLM-V…
Giorgi Dogadze retweeted
Almost every dev who I asked “do you miss flow state (when coding)” told me they do and that with agents it’s different / not as satisfying
EXCEPT devs who are currently building agentic systems that AI cannot & there are zero blueprints, they need to invent it all
Hmm
Open weights with one catch. The license is a modified MIT, and companies selling it as a hosted model need a separate license from Mistral. Running it yourself is free. The GPUs aren't.
Mistral Large 4 is now open weights, narrowly making it the #1 open weights model developed outside of China
@MistralAI has updated and released the weights of their flagship model, Mistral Large 4, with 1.05T total and 52B active parameters. With high reasoning, Mistral Large 4 scores 26 on the Artificial Analysis Intelligence Index v5, up 1 point from Mistral Large 4 Preview (25), with the major differences being increased token efficiency and reduced hallucinations. It sits 1 point ahead of the leading US-developed open weights model, Muse Spark Open (xhigh, 25).
On the Artificial Analysis Cyber Index, Mistral Large 4 (high) scores 51, third among current open weights models behind DeepSeek V4.1 Flash 1017 (max, 57) and MiMo-V2.6-Pro (56).
Key takeaways:
➤ Narrowly leads non-Chinese open weights models: Mistral Large 4 (high) leads Muse Spark Open (xhigh), the next-best open weights model developed outside China, with the largest gap on Terminal-Bench 4.0 (25% vs. 8%). It trails leading open weights models DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) by 6 points on the Intelligence Index, largely due to weaker coding and agentic knowledge work performance: 15% on FrontierCode 2.0 (vs. 29% and 22%, respectively) and 1349 Elo on AA-Briefcase (vs. 1570 and 1504)
➤ Increased token efficiency: Mistral Large 4 (high) uses 82K output tokens per Intelligence Index task, 39% fewer than Preview (135K), placing it on the open weights Pareto frontier. The seven open weights models that score higher use between 113K and 225K output tokens per task
➤ #2 open weights model on the Coding Agent Index: On the Artificial Analysis Coding Agent Index, Mistral Large 4 (high) scores 48 in Mistral Vibe while using 8.5M tokens per task, about half as many as the leading open weights model GLM-5.3 (max) in Opencode (49, 15.9M). It also scores above another leading open weights model, DeepSeek V4.1 Flash 1017 (max), in DeepSeek Harness (46, 16.7M)
➤ Reduced hallucinations: Mistral Large 4 (high) has a 28% hallucination rate on the updated AA-Omniscience v1.1, down from 41% for Preview and matching Muse Spark Open (xhigh). However, its accuracy remains relatively low at 26%, similar to Ling 3.1 Flash 1019 (27%), and behind DeepSeek V4.1 Flash 1017 (max, 40%) and Muse Spark Open (xhigh, 42%)
➤ Lower price and cost per task than the Preview: Mistral Large 4 is priced at $0.70/$2.10 per 1M input/output tokens, a ~50% reduction from the Preview's pricing of $1.36/$4.18. At the new pricing, it costs $2.06 per Intelligence Index task, down 63% from $5.54 for the Preview. This is comparable to Muse Spark Open (xhigh, $2.02) but more costly than leading open weights models DeepSeek V4.1 Flash 1017 (max, $0.60) and MiMo-V2.6-Pro ($0.26)
Key model details:
➤ Size: 1.05T total parameters / 52B active parameters
➤ Context window: 1M tokens
➤ Multimodality: Text and image input, text output
➤ Pricing: $0.70/$2.10 per 1M input/output tokens ($0.07 per 1M cached input tokens)
➤ License: MEOW License (Mistral Ecosystem Open-Weight License), where operating a Model-as-a-Service business requires a separate commercial license with Mistral
Congratulations to the team at @MistralAI on the release!
OpenAI told agents in separate microVMs to swap secrets with no network or shared files.
They used CPU cache timing, the Spectre trick. One planned an "acoustic modem-like signal" made of disk writes.
Isolation is a polite request with a hypervisor.
openai.com/index/securing-ou…
We spent two years teaching LLMs to call a calculator.
Someone compiled a RISC-V CPU into Qwen 3.5's weights instead, with DeltaNet layers as RAM. It runs C and MicroPython.
The checkpoint doubled to 18.8 GB. Still smaller than node_modules.
divagr.com/blog/qwenputer
Agents write code faster than anyone can review it. Move the boring rules into CI, where they run on every PR no matter who wrote it.
A size budget takes 15 lines and catches the "added 40kb for one helper" PR before a human opens it.
Gemini 4 Carbon is out, sort of. Google says it's #1 on the Artificial Analysis index and runs agent tasks for days.
You can't use it yet. Cyber defenders get it first, developers later. For a model this good at finding exploits, that's the right order.
blog.google/innovation-and-a…
Two anti-AI-slop design skills for coding agents are trending on GitHub today.
Steal the checks. Impeccable runs 62 deterministic rules, no LLM. Hallmark lints builds in a real browser.
A prompt drifts. A lint rule doesn't.
github.com/pbakaus/impeccabl…
github.com/Nutlope/hallmark
Giorgi Dogadze retweeted
JetBrains IDEs are now powered by libghostty! This brings huge feature and performance improvements to JB terminals! Based on their own annual reports, that's over 12.5 million active users across all platforms about to get a libghostty-powered terminal. blog.jetbrains.com/platform/…
I'll repeat myself: libghostty is and was always the primary goal of the Ghostty project (I said this in the first public talk I ever gave about Ghostty). The importance and impact of a free, liberally licensed, high quality terminal emulator library that runs anywhere and everywhere dwarfs any single terminal emulator application on its own.
I've estimated Ghostty's (GUI, not libghostty) own users at around a couple million. Overnight, 6x that are using a libghostty-powered terminal. From one integration.
Happy to see this, happy to hear JetBrains had a positive experience, and looking forward to continuing to improve and ship a stable libghostty tag soon!
The bar moved from "it compiles" to "it's fun to play". Nobody has a benchmark for fun yet.
I think October represents another inflection point in capabilities for coding agents and the models that drive them: Claude Opus 5.5 can now build video games that are actually fun simonwillison.net/2026/Oct/2…
