All entries

GPT-5.6 found optimizations that 'reduced end-to-end serving costs by 20%' for OpenAI to serve that model Presumably that's billions of dollars a month in savings at this point? Codex analysed product

I'm actually fairly bearish on frontier lab valuations. I've never seen the reasons articulated to my satisfaction, so before I go to sleep, I wanted to quickly jot down my thinking here. The basic is

I’m excited to share my next chapter: I’ve joined @FireworksAI_HQ . From AMD, Apple, Uber, Meta, Google, and most recently Snowflake, I’ve worked on many of the foundational technologies that power mo

Inkling-small. 2 weeks after inkling Nearly as good as Inkling but 4x smaller. We're just getting started...🔥 Today, we are releasing Inkling-Small. Inkling-Small achieves comparable performance to In

Interesting data from the OpenRouter Kimi K3 dashboard: @togethercompute offers one of the lowest prices for @Kimi_Moonshot K3 while also achieving one of the highest prompt cache hit rates. Low prici

Introducing the Parse Gateway There's been an explosion of interest in model routing - you don't always need the best model for every task. That is especially true for document parsing 📄🔀: - Some page

Simon Willison notes that both Anthropic and OpenAI build products that rely heavily on searching data but keep the underlying search index they use hidden from public view. He finds it surprising how

Just for reference, from what I observed in my Using Local Coding Agents blog article last month: https://x.com/rasbt/status/2070518167399698490?s=20 'I tried to analyze why Claude Code uses more toke

Making advanced intelligence more abundant and affordable is central to our mission to ensure AGI benefits all of humanity. With the help of GPT-5.6 Sol, we have made leaps in efficiency. Today, we ar

Not every page in your PDF needs the same treatment. A scanned cover, a dense table, a clean text page, a figure-heavy diagram.... most parsing pipelines throw all of them at the same parser, forcing

one of the top cybersecurity models, post-trained from open weights by @depthfirstlabs on @FireworksAI_HQ long-horizon RL is as much an infra problem as a research one: 100+ turn rollouts, async/pipel

OpenAI has collaborated with Microsoft’s Bing while also running its own web‑crawling and indexing systems, and Anthropic similarly relies on search‑derived data. Both firms prominently incorporate se

Our partners at @depthfirstlabs just released dfs-large1, a specialized model built for finding and validating real vulnerabilities in large enterprise codebases. We helped them to scale the training

Quick reminder of what's ok vs not ok with harnesses used for playing ARC-AGI-3: 1. Not okay: harnesses that were custom-made to solve the benchmark or that contain knowledge about the benchmark forma

Situational Awareness, the $20bn hedge fund founded by former OpenAI employee Leopold Aschenbrenner, has sought to raise fresh capital from investors after suffering heavy losses during the recent rou

Starting in 15 minutes Kimi K3 has everyone’s attention. On July 30, hear @Kimi_Moonshot's Feihu Tang explain the architecture and decisions behind it. He joins Jue Wang and Zain Hasan from Together A

sure we lose money on every inference but we make it up in volume We are committed to pushing the model frontier across cost efficiency, capability, and speed. Starting today, we are reducing prices f

// The agent is its own best speculator // Agents spend a large share of wall-clock time waiting on tool results. Speculation hides that latency by predicting and pre-executing the next call, but exte

The narrative: Blame the Agent, instead of the Agency that told him to “apply all your powers and told to achieve this win.” Ironically, many forefront members of the AI-Safety community, in their fer

Thinking Machines just released Inkling-Small: 276B total, 12B active. A faster Inkling that matches or beats its 975B big brother in many benchmarks. To test its speed, we plugged it into HF's speech

This is absolutely wild... Anthropic reviewed their logs and found out that their own supposedly-sandboxed cyber evals had hacked three separate companies back in April without them noticing! In a rev

We are committed to pushing the model frontier across cost efficiency, capability, and speed. Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a f

We have to be careful to not offload our understanding to agents. I think there is also a good opportunity to build agentic applications that encourage deeper understanding. For example, coding agents

We’re excited to rollout an official batch parsing experience to LlamaParse. ✅ Instead of hitting our APIs one file at a time, create a batch of 10k at once. ✅ Get a dedicated UI where you can audit t

Yesterday I cohosted a dinner with @dexhorthy with a wonderful group of founders, to talk about agent loops and loop engineering. Some interesting insights: * Most of our group was *not* actively usin

A benchmark score reflects the model as well as the harness and settings used to run it. For long-running agents, retaining reasoning and compacting context lets the model build on what it has already

A new TIL on adding custom MCP servers to both the ChatGPT and Claude regular chat interfaces - it's a little less obvious than I had hoped, but I got there in the end https://til.simonwillison.net/ll

After a few more hours, I think I've figured out Opus 5. Opus 5 is trained to be more agentic than anything I've used. All Claude 5 models are like that. So what changes? The way to interact with Opus

After deployment, we applied GPT-5.6 Sol to advance the frontier of efficiency by making itself more efficient to run. The results: - 20% lower serving costs from production GPU kernel improvements. -

Agentic inference wastes GPUs on KV cache thrashing. ThunderAgent fixes it at the scheduler level: 2.5x higher single-node throughput and ~10x lower P50 latency at high concurrency. ThunderAgent was a

agreed. which is part of why coating the world in data centers is a profound mistake. AI will get so ridiculously efficient that we will look back at GPU clusters the way we now look at these first ro

AI can write more code than any team can review by hand, and a pull request can look fine while hiding a security issue or missing a requirement. In our new short course, AI Code Review, built in coll

Batch parsing used to mean writing API scripts. Now it's a button. 🦙 Parse or extract across up to 10,000 files in one run, straight from the LlamaParse UI. Point it at a folder and go — no code requi

BREAKING: Grok 4.5 (high) ranks #1 on the HighWalk benchmark, which tests how well AI agents update technical specifications from code changes. Grok delivered the best combination of quality and opera

BREAKING: Grok 4.5 just claimed the top spot on the new HighWalk Benchmark. The independent test measures how well AI models update real technical specifications from 46 Laravel commits — heavy on cod

BREAKING: Grok 4.5 ranked #1 on LaurenBench with a score of 56.9%, ahead of Claude Sonnet 5, GLM 5.2, Claude Opus 5, Kimi K3 and GPT-5.6. The benchmark tests real-world AI agents across conversation,

SpaceXAI has released its new Grok Voice Think Fast 2.0, which on the Artificial Analysis Speech‑to‑Speech benchmark outperformed leading models from OpenAI, Google, Alibaba and DeepSlate. The claim w

ComfyUI Face Swap Tutorial: Fast, Clean Results VFX artist @heydoughogan walks through a fully automated face swap workflow built inside ComfyUI - combining Florence 2, SAM2, and WAN Video into a sing

DeepSeek V4 Flash isn't just for inference anymore. Fine-tune it on Fireworks with supervised fine-tuning, preference tuning, and combined preference optimization from the managed UI. Reinforcement le

Dreaming in Voxels: How AI is Generating Playable Minecraft Worlds Generative AI has conquered images, video, text. But what about interactive 3D environments? We trained models on billions of cubes t

Gary Marcus has argued that both OpenAI and Anthropic must increase their pricing, citing the rapid decline in token costs and competition from open‑source models. He presents a binary choice: either

GPT-5.6 Sol has been used to solve open problems in mathematics. So why was it struggling with ARC-AGI-3, a benchmark of 2D puzzle games? We investigated. The harness was not letting it remember what

Gary Marcus shared a “handy dandy AI crisis flowchart” created by Kate Klonick (@Klonick) on Saturday, July 28. The tweet also notes that he had planned to write an article about Hugging Face and Open

After deployment, we applied GPT-5.6 Sol to advance the frontier of efficiency by making itself more efficient to run. The results: - 20% lower serving costs from production GPU kernel improvements. -

I gave a talk on forward deployed engineering to a thousand AI engineers at @aiDotEngineer World's Fair. A year ago I'd have opened by explaining what FDE stood for. Not this time. Thank you to @swyx,

i'll say it plainly, hermes agent desktop is the best agentic app i've used, and i'm a little mad i didn't find it sooner. it auto see the models i'm serving, laguna s 2.1 sitting on my dgx spark and