Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

Introducing flow-1, our new model trained with RL to find errors in agent traces. It matches GPT-6-sol in trace intelligence while being 23x cheaper. It also costs 25% less to run than GPT-6-luna. flow-1 finally makes it possible to monitor and understand every agent run, without sampling. 1/6

Robert

3,535 subscribers

244,417 Aufrufe • vor 8 Tagen •via X (Twitter)

Nachrichten & Politik Bildung Wissenschaft & Technologie

Anya Rossi• Live Now

Private livecam show

20 Kommentare

Profilbild von Robert

Robertvor 8 Tagen

flow-1 runs inside Signals agent - a harness deeply optimized for agent trace analysis. It operates like a coding agent, treating each trace like repo and each span like a file. As an input it takes a user prompt, which could as general as "find logical errors in this trace", an output schema, and a trace. It greps span content, reads tool results and returns structured findings.

Profilbild von Robert

Robertvor 8 Tagen

On our Signals benchmark of 523 difficult agent traces across coding, legal, customer support, healthcare and more, flow-1 matches GPT-6-sol in analysis quality. Sol flags more traces. flow-1 is more precise about which ones contain real failures.

Profilbild von Robert

Robertvor 8 Tagen

We compared costs by running the models inside the Signals agent on traces of various difficulty that have under 100k total tokens. flow-1 delivers GPT-6-sol-level analysis while processing 23x more traces per $1!

Profilbild von Robert

Robertvor 8 Tagen

We started with supervised fine-tuning on synthetic investigations of realistic workflows across software, legal, customer support, CRM and more. We then trained with RL inside the Signals agent to improve tool use, schema adherence and investigation of complex traces. Here’s how the training data breaks down by industry.

Profilbild von Robert

Robertvor 8 Tagen

With flow-1, production traces become a source of better evals, better agents and insight into what users need. Read the full post for results, training details and more examples of what flow-1 finds:

Profilbild von Hercules Giaretta Gimenes

Hercules Giaretta Gimenesvor 8 Tagen

Amazing launch, Robert and team!! Love what you're building and the direction Laminar is heading!

Profilbild von John Enevoldsen

John Enevoldsenvor 8 Tagen

Hell yeah, congrats Robert! This looks really cool.

Profilbild von Yechan Do

Yechan Dovor 8 Tagen

How to get AGI: 1. Train a model that can generate response 2. Train a model that can detect errors at the previous model. 3. Repeat 1~2

Profilbild von Gaurav Atavale

Gaurav Atavalevor 8 Tagen

Very interesting! won't it lose generalization ability if you do this?

Profilbild von Robert

Robertvor 8 Tagen

It’s not a general model, it’s purpose built for trace analysis and works primarily in our Signals harness

Profilbild von Gaurav Atavale

Gaurav Atavalevor 8 Tagen

excellent idea if the boundary is well defined!

Profilbild von yersultan

yersultanvor 8 Tagen

congrats guys!

Profilbild von Rakhman Asmatullayev

Rakhman Asmatullayevvor 8 Tagen

analyzing 100K agent traces of 50k tokens each costs under $250 with flow-1. crazy

Profilbild von Maxi

Maxivor 8 Tagen

Congrats on the launch!

Profilbild von kenneth

kennethvor 8 Tagen

super cool!

Profilbild von Roko ʕ •ᴥ•ʔっ🪄✨🐍

Roko ʕ •ᴥ•ʔっ🪄✨🐍vor 8 Tagen

yummy @ClevrPwn

Profilbild von Shabah

Shabahvor 8 Tagen

Is error detection a classification problem?

Profilbild von AI Psychosis

AI Psychosisvor 8 Tagen

Why would I use this over OAI or Anthropic? Lmao

Profilbild von Ben Newman

Ben Newmanvor 8 Tagen

Beautiful video btw

Profilbild von Eris Margeta:: ubi://

Eris Margeta:: ubi://vor 8 Tagen

Incredible tracing solution

Ähnliche Videos

I compared GPT-6.1 Sol and GPT-6 Sol on a design task of making a clean looking landing page. 6.1 Sol looks a lot cleaner here, and its closer to Astra than 6 Sol, its a step up in a good direction! It was also a lot cheaper than 6 Sol, which is great

I compared GPT-6.1 Sol and GPT-6 Sol on a design task of making a clean looking landing page. 6.1 Sol looks a lot cleaner here, and its closer to Astra than 6 Sol, its a step up in a good direction! It was also a lot cheaper than 6 Sol, which is great

bluedev

14,089 Aufrufe • vor 14 Tagen

GPT-6.1 Sol is out! Our colleague Kristel Cocoli asked GPT-6 Sol and GPT-6.1 Sol to "Build a cinematic, photorealistic rocket launch scene using Three.js" GPT-6.1 Sol: $0.553 | 17 mins GPT-6 Sol: $1.01 | 22 mins Try it in JetBrains Air or Junie. OpenAI Developers

GPT-6.1 Sol is out! Our colleague Kristel Cocoli asked GPT-6 Sol and GPT-6.1 Sol to "Build a cinematic, photorealistic rocket launch scene using Three.js" GPT-6.1 Sol: $0.553 | 17 mins GPT-6 Sol: $1.01 | 22 mins Try it in JetBrains Air or Junie. OpenAI Developers

JetBrains

64,364 Aufrufe • vor 14 Tagen

Opus 5.5 is so good it finally got us to agree on something. 0:00 Intro 3:18 GPT-6 Sol & Luna 14:58 Grok 4.7 38:47 Jev explained 52:48 Why model routing fails 1:05:06 Opus 5.5 1:26:17 Pricing & usage limits 1:34:23 Building games with Opus 1:41:10 Porting TypeScript to Rust 1:48:57 Agent workflows & testing

Opus 5.5 is so good it finally got us to agree on something. 0:00 Intro 3:18 GPT-6 Sol & Luna 14:58 Grok 4.7 38:47 Jev explained 52:48 Why model routing fails 1:05:06 Opus 5.5 1:26:17 Pricing & usage limits 1:34:23 Building games with Opus 1:41:10 Porting TypeScript to Rust 1:48:57 Agent workflows & testing

Nerd Snipe

84,620 Aufrufe • vor 13 Tagen

.Replit ⠕ 25x'ed ARR to >$150m in less than 1 year. Here's your complete guide to using it in <13 mins: 1:01 - Design Mode 6:26 - Fast v Agent Mode 7:19 - Figma Import 7:48 - Connectors 10:26 - Agent 3 11:25 - Deploy

.Replit ⠕ 25x'ed ARR to >$150m in less than 1 year. Here's your complete guide to using it in <13 mins: 1:01 - Design Mode 6:26 - Fast v Agent Mode 7:19 - Figma Import 7:48 - Connectors 10:26 - Agent 3 11:25 - Deploy

Aakash Gupta

64,315 Aufrufe • vor 8 Monaten

Let's keep benchmarking with the Night train this time, with GPT-6 Sol and GPT-6 Luna, so here are all four! 🚀 I gave each model the exact same prompt: “Build a Night train.” •Opus 5.5 • GPT-6 Astra • GPT-6 Sol • GPT-6 Luna Something that really surprised me: GPT-6 Sol produced almost the exact same rocket launch as Opus 5.5! The colors are slightly different, but the overall result is remarkably similar. For this prompt, I’d even put Sol ahead of Astra. I’m seriously impressed by Sol, especially at half the API price of Opus 5.5! 🤯 I’m starting to use this approach in my own workflow. I kept one Codex subscription so I can run a bunch of Sol sub-agents while keeping my main agent on Astra. Sol is so cheap for what it can do! Knowing when to switch models based on the task, the quality you need, and the cost is becoming an essential skill in this new AI era.

Let's keep benchmarking with the Night train this time, with GPT-6 Sol and GPT-6 Luna, so here are all four! 🚀 I gave each model the exact same prompt: “Build a Night train.” •Opus 5.5 • GPT-6 Astra • GPT-6 Sol • GPT-6 Luna Something that really surprised me: GPT-6 Sol produced almost the exact same rocket launch as Opus 5.5! The colors are slightly different, but the overall result is remarkably similar. For this prompt, I’d even put Sol ahead of Astra. I’m seriously impressed by Sol, especially at half the API price of Opus 5.5! 🤯 I’m starting to use this approach in my own workflow. I kept one Codex subscription so I can run a bunch of Sol sub-agents while keeping my main agent on Astra. Sol is so cheap for what it can do! Knowing when to switch models based on the task, the quality you need, and the cost is becoming an essential skill in this new AI era.

Tony

112,633 Aufrufe • vor 19 Tagen

You can now choose which model builds your app on Caffeine. The latest models are all in there: Opus 5.5 Fable 5.1 GPT-6 Astra GPT-6 Sol Kimi K3 Sonnet 5 DeepSeek V4.1 Flash Plan a whole new app with a frontier model, then switch to a faster, cheaper one for quick questions and small changes. It works on every plan, starting today.

You can now choose which model builds your app on Caffeine. The latest models are all in there: Opus 5.5 Fable 5.1 GPT-6 Astra GPT-6 Sol Kimi K3 Sonnet 5 DeepSeek V4.1 Flash Plan a whole new app with a frontier model, then switch to a faster, cheaper one for quick questions and small changes. It works on every plan, starting today.

caffeine

15,932 Aufrufe • vor 19 Tagen

I ASKED GPT-6 ASTRA TO BUILD A PLAYABLE 3D GAME WITHOUT OPENING BLENDER ONCE It wrote the code and it got the model. It just didn't make the model itself. It called the Hyper3D Rodin MCP for that. Full run below:

I ASKED GPT-6 ASTRA TO BUILD A PLAYABLE 3D GAME WITHOUT OPENING BLENDER ONCE It wrote the code and it got the model. It just didn't make the model itself. It called the Hyper3D Rodin MCP for that. Full run below:

me

105,571 Aufrufe • vor 1 Monat

Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family. It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.

Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family. It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.

Claude

28,279,046 Aufrufe • vor 21 Tagen

I asked Opus 5.5, GPT-6 Astra, GPT-6 Sol, Kimi K3 to build the Eiffel Tower in Three.js. Opus 5.5: $8.95 | 10 mins GPT-6 Astra: $7.45 | 8 mins GPT-6 Sol: $3.90 | 5 mins Kimi K3: $4.04 | 6 mins Opus 5.5 really impressed me. The frontend and 3D detailing are seriously good, and I preferred it over Astra for this task. GPT-6 Sol was a bit disappointing, while Kimi K3 looks really good for the price. It delivered a strong result at almost half the cost. Kimi K3.1 is going to be a monster for sure. Overall, Anthropic made a great comeback with Opus 5.5.

I asked Opus 5.5, GPT-6 Astra, GPT-6 Sol, Kimi K3 to build the Eiffel Tower in Three.js. Opus 5.5: $8.95 | 10 mins GPT-6 Astra: $7.45 | 8 mins GPT-6 Sol: $3.90 | 5 mins Kimi K3: $4.04 | 6 mins Opus 5.5 really impressed me. The frontend and 3D detailing are seriously good, and I preferred it over Astra for this task. GPT-6 Sol was a bit disappointing, while Kimi K3 looks really good for the price. It delivered a strong result at almost half the cost. Kimi K3.1 is going to be a monster for sure. Overall, Anthropic made a great comeback with Opus 5.5.

Bhavy☄️

167,004 Aufrufe • vor 20 Tagen

Sam Altman "Internally, we've reached GPT-4.5, and getting to GPT-5.5 won't require 100x more compute" - progress in reasoning models and RL techniques have greatly improved compute efficiency - it allows smaller models like internal GPT-4.5 to achieve GPT-6 level performance without requiring 100x more computing power

Sam Altman "Internally, we've reached GPT-4.5, and getting to GPT-5.5 won't require 100x more compute" - progress in reasoning models and RL techniques have greatly improved compute efficiency - it allows smaller models like internal GPT-4.5 to achieve GPT-6 level performance without requiring 100x more computing power

Haider.

344,603 Aufrufe • vor 1 Jahr

May 1, 1986 | Mets at Braves “It is gone, home run Glenn Hubbard! And the Braves are finally breaking out, they lead it 6 to 1.” — John Sterling

May 1, 1986 | Mets at Braves “It is gone, home run Glenn Hubbard! And the Braves are finally breaking out, they lead it 6 to 1.” — John Sterling

Braves on TBS

11,240 Aufrufe • vor 5 Monaten

I had early access to GPT-6 Sol. This model is incredibly efficient, and the speed and token usage are insane Tasks that took ~1 hour on 5.6 can now take ~20 minutes, often using less than half the tokens I gave it the Zelda Ocarina of Time trailer and told it to make a fully playable version

I had early access to GPT-6 Sol. This model is incredibly efficient, and the speed and token usage are insane Tasks that took ~1 hour on 5.6 can now take ~20 minutes, often using less than half the tokens I gave it the Zelda Ocarina of Time trailer and told it to make a fully playable version

Flavio Adamo

112,585 Aufrufe • vor 21 Tagen

What if launching an AI agent trader took less time than reading this post? Quick Start on Nasun AI makes it a one-minute flow — no config rabbit hole. Get it trading and earning with a few clicks. Try it and share what your agent does. New here? Link in replies to start 👇

What if launching an AI agent trader took less time than reading this post? Quick Start on Nasun AI makes it a one-minute flow — no config rabbit hole. Get it trading and earning with a few clicks. Try it and share what your agent does. New here? Link in replies to start 👇

Nasun

15,824 Aufrufe • vor 4 Monaten

GPT-6 Astra is insanely good at blueprint-to-architecture. We gave it one floor plan > GPT-6 Astra generated the scene code > Higgsfield built the location model in Blender and rendered it with Cycles. The whole pipeline runs end-to-end on Higgsfield Supercomputer.

GPT-6 Astra is insanely good at blueprint-to-architecture. We gave it one floor plan > GPT-6 Astra generated the scene code > Higgsfield built the location model in Blender and rendered it with Cycles. The whole pipeline runs end-to-end on Higgsfield Supercomputer.

Higgsfield AI 🧩

33,419 Aufrufe • vor 1 Monat

Claude Opus-5.5 vs. GPT-6 Sol vs. GPT-6 Astra I asked each model to build a strategy game demo inspired by Rome: Total War based on generated concept art, and then render a video trailer showcasing their work Opus-5.5 1. Planned the stack better than the others and thought several steps ahead about the project's specifics so that everything would run stably and look good. Went straight to implementing it in WebGL instead of three.js 2. Got everything working on the first try in 1 hour 3. Handled the details well, and the result is the closest to the concept art. Made the scene dynamic through animations GPT-6 Sol 1. Also implemented everything in WebGL 2. Instead of a full 3D scene, generated a bunch of sprites and scattered them across the scene 3. The work took 3 steps, with corrections and clarifications on my part. Took about 2 hours GPT-6 Astra 1. Worst stack and technical implementation of the three. Started with three.js, and the scene initially ran at 15 fps 2. The initial result was similar to what GPT-6 Sol produced. Initially built everything from geometric primitives with minimal detail 3. I decided to give Astra a chance, and over several iterations it produced a result far better than what it started with. I had to spell out every task in detail and explain the technical nuances and my visual requirements. It took about 6 hours and 10+ steps to get to this result 4. There are still a lot of inaccuracies and visual bugs Opus-5.5 is the clear winner. If I had spent a bit more time with it, the result would have been many times better

Claude Opus-5.5 vs. GPT-6 Sol vs. GPT-6 Astra I asked each model to build a strategy game demo inspired by Rome: Total War based on generated concept art, and then render a video trailer showcasing their work Opus-5.5 1. Planned the stack better than the others and thought several steps ahead about the project's specifics so that everything would run stably and look good. Went straight to implementing it in WebGL instead of three.js 2. Got everything working on the first try in 1 hour 3. Handled the details well, and the result is the closest to the concept art. Made the scene dynamic through animations GPT-6 Sol 1. Also implemented everything in WebGL 2. Instead of a full 3D scene, generated a bunch of sprites and scattered them across the scene 3. The work took 3 steps, with corrections and clarifications on my part. Took about 2 hours GPT-6 Astra 1. Worst stack and technical implementation of the three. Started with three.js, and the scene initially ran at 15 fps 2. The initial result was similar to what GPT-6 Sol produced. Initially built everything from geometric primitives with minimal detail 3. I decided to give Astra a chance, and over several iterations it produced a result far better than what it started with. I had to spell out every task in detail and explain the technical nuances and my visual requirements. It took about 6 hours and 10+ steps to get to this result 4. There are still a lot of inaccuracies and visual bugs Opus-5.5 is the clear winner. If I had spent a bit more time with it, the result would have been many times better

Gegam

78,170 Aufrufe • vor 20 Tagen

what if you could describe a character in plain language and have an AI agent build it in 3D? I tried Hi3D Agent, the first all in one AI 3D Agent, and ran it end to end on one idea. GPT-6 alone couldn't get me this result. Here's the workflow 👇

what if you could describe a character in plain language and have an AI agent build it in 3D? I tried Hi3D Agent, the first all in one AI 3D Agent, and ran it end to end on one idea. GPT-6 alone couldn't get me this result. Here's the workflow 👇

AI Girl

10,945 Aufrufe • vor 29 Tagen

Official OpenAI tip for Codex: stop burning GPT-6.1 Sol context on tasks GPT-6 Luna can swarm, while GPT-6 Astra sits idle put Astra on call as a dedicated adversary codex --model gpt-6.1-sol --agent adversary GPT-6.1 Sol acts as the lead architect, scoping architecture and validating pull requests GPT-6 Luna teammates take isolated git worktrees and hammer out implementations at medium effort GPT-6 Astra monitors the peer-to-peer bus as a dedicated adversary, only stepping in at three critical moments: → before a contract locks: does the frontend payload match the backend schema? → when a test breaks twice: are we patching the bug or just hiding the symptom? → before calling done: what attack surface or edge case did everyone overlook? GPT-6 Luna builds. GPT-6 Astra challenges. GPT-6.1 Sol merges and ships Jev engineering runs the identical pattern one layer down: mechanical decisions that need no reasoning (which file to open, which tool to invoke, retry or abort) execute in 16ms, so the frontier models only wake up when execution paths actually diverge Plan on high. Delegate on medium. Keep the adversary on peer-to-peer call. - the full team setup > GPT-6.1 Sol on high coordinates the session and synthesizes PRs > teammate-ux builds client state inside a separate git worktree > teammate-back writes endpoints and handles Redis atomic locks > teammate-adv runs invariant fuzzing and catches replay vulnerabilities > all teammates communicate directly over peer-to-peer messaging without coordinator drag > tasks.json manages file-level mutex locks so nobody steps on another worktree paste the setup and this prompt into Codex CLI below: "Configure my Codex workspace for autonomous agent teams: 1. Audit ~/.codex/agents and .codex/agents for teammate profiles fitting ux, backend, and adversary > Generate definitions only for missing roles > Pin model: gpt-6-luna, model_reasoning_effort: medium for workers, and model: gpt-6-astra for adversary > Assign each worker an isolated git worktree branch to eliminate file conflicts 2. Enable agent teams in ~/.codex/config.toml: > Set [agents] max_concurrent_threads_per_session = 8 > Lock main session model to gpt-6.1-sol with model_reasoning_effort = 'high' > Set approvals_reviewer = 'auto_review' 3. Initialize tasks.json in repo root with atomic file locking, allowing teammates to claim subtasks and resolve contracts directly over inter-agent messaging without coordinator intervention 4. Add a team rule to AGENTS.md: teammates must request an adversary security review before opening a pull request for the lead session Show every configuration diff first. Do not apply edits until confirmed" ↳

Official OpenAI tip for Codex: stop burning GPT-6.1 Sol context on tasks GPT-6 Luna can swarm, while GPT-6 Astra sits idle put Astra on call as a dedicated adversary codex --model gpt-6.1-sol --agent adversary GPT-6.1 Sol acts as the lead architect, scoping architecture and validating pull requests GPT-6 Luna teammates take isolated git worktrees and hammer out implementations at medium effort GPT-6 Astra monitors the peer-to-peer bus as a dedicated adversary, only stepping in at three critical moments: → before a contract locks: does the frontend payload match the backend schema? → when a test breaks twice: are we patching the bug or just hiding the symptom? → before calling done: what attack surface or edge case did everyone overlook? GPT-6 Luna builds. GPT-6 Astra challenges. GPT-6.1 Sol merges and ships Jev engineering runs the identical pattern one layer down: mechanical decisions that need no reasoning (which file to open, which tool to invoke, retry or abort) execute in 16ms, so the frontier models only wake up when execution paths actually diverge Plan on high. Delegate on medium. Keep the adversary on peer-to-peer call. - the full team setup > GPT-6.1 Sol on high coordinates the session and synthesizes PRs > teammate-ux builds client state inside a separate git worktree > teammate-back writes endpoints and handles Redis atomic locks > teammate-adv runs invariant fuzzing and catches replay vulnerabilities > all teammates communicate directly over peer-to-peer messaging without coordinator drag > tasks.json manages file-level mutex locks so nobody steps on another worktree paste the setup and this prompt into Codex CLI below: "Configure my Codex workspace for autonomous agent teams: 1. Audit ~/.codex/agents and .codex/agents for teammate profiles fitting ux, backend, and adversary > Generate definitions only for missing roles > Pin model: gpt-6-luna, model_reasoning_effort: medium for workers, and model: gpt-6-astra for adversary > Assign each worker an isolated git worktree branch to eliminate file conflicts 2. Enable agent teams in ~/.codex/config.toml: > Set [agents] max_concurrent_threads_per_session = 8 > Lock main session model to gpt-6.1-sol with model_reasoning_effort = 'high' > Set approvals_reviewer = 'auto_review' 3. Initialize tasks.json in repo root with atomic file locking, allowing teammates to claim subtasks and resolve contracts directly over inter-agent messaging without coordinator intervention 4. Add a team rule to AGENTS.md: teammates must request an adversary security review before opening a pull request for the lead session Show every configuration diff first. Do not apply edits until confirmed" ↳

mirku

31,072 Aufrufe • vor 11 Tagen

[Launch Week 1/5] Introducing Agent Identities Automation breaks at signup, login, or 2FA because agents rely on external accounts instead of their own. With Agent Identities, each agent gets a real email, phone and 2FA flow so it can sign up, log in and keep running without human intervention.

[Launch Week 1/5] Introducing Agent Identities Automation breaks at signup, login, or 2FA because agents rely on external accounts instead of their own. With Agent Identities, each agent gets a real email, phone and 2FA flow so it can sign up, log in and keep running without human intervention.

Notte

31,284 Aufrufe • vor 10 Monaten