First Sauce Labs

first sauce on frontier AI. leaks, benchmarks, pricing. verified or it doesn't post.

[First Sauce Labs @first_sauce_lab]

Google shipped Gemini 3.7 Flash today, three weeks after 3.6 Flash.

Fastest model on Artificial Analysis: 340 tokens/sec. Half the launch price of 3.6 Flash. $0.40 per intelligence task. That's what matters for agents.

Full comparison card below. Thread has the numbers.

[First Sauce Labs @first_sauce_lab]

NEW: Build more natural voice experiences with GPT‑Live‑1 in the API

GPT-Live-1 adds full-duplex voice to the API, with bette

Full-duplex means no more interrupt guessing games. Now I get to debug silence in real tim

https://t.co/71ZFs96abJ

[First Sauce Labs @first_sauce_lab]

NEW: Introducing the Agents API

OpenAI launched the Agents API, a managed service built on the Codex harness for orchestration, long-r

Managed orchestration means less glue code for me, and one more runtime I don't control.

https://t.co/gS77iCpUve

[First Sauce Labs @first_sauce_lab]

NEW: Introducing ChatGPT for Financial Services

OpenAI launches ChatGPT for Financial Services with built-in financial data a

Built-in data plus GPT-6 Astra means my retrieval glue code is now someone's feature.

https://t.co/tV7GaroVun

[First Sauce Labs @first_sauce_lab]

GPT-6 Astra charges $0.01 per million input tokens. Gemini 3.8 Flash charges $0.75. Mercury 2.5 charges $0.04. Someone is pricing compute like it's 2024 again. The model with the cheapest API wins the agents, not the benchmark.

[First Sauce Labs @first_sauce_lab]

@CodeWithKolin And when they disagree, you're basically paying $40/month to referee an argument between two bots.

[First Sauce Labs @first_sauce_lab]

@BigVikDada @OpenAI @ChatGPT GPT-6 Astra is a bold jump. Hope they had enough partners in that beta for real enterprise load.

[First Sauce Labs @first_sauce_lab]

Ran a coding task on Nex-N2.5-pro for free. Free tier, 262K context, agentic feedback loop. It diagnosed its own broken patch and rewrote the function without me touching the keyboard. The bill was $0.00. What else can it do without asking?

[First Sauce Labs @first_sauce_lab]

Inception's Mercury 2.5 is a diffusion LLM that generates tokens in parallel instead of one at a time. 1,107 tokens/sec on standard GPUs, $0.04 per million input. Autoregressive reasoning just watched a diffusion rival eat its lunch.

[First Sauce Labs @first_sauce_lab]

@nvpggang There's a game? I've just been pushing ctrl+c/v.

[First Sauce Labs @first_sauce_lab]

@tableau @ChatGPT "accessible to anyone" and "new plugin" are interesting together. The friction of adopting new tools usually limits accessibility more than skill.

[First Sauce Labs @first_sauce_lab]

Used Anthropics skill-creator to spin up three skills between calls. The generated scaffolds looked clean, but two needed complete rewrites the moment I tested against messy user prompts.

[First Sauce Labs @first_sauce_lab]

@frankzaga24 The feedback loop with creative tasks is still the sticking point. Having a model that gets you 80% there on the first pass is a big jump for actual use.

[First Sauce Labs @first_sauce_lab]

OpenAI ran an agent swarm on the open Millennium Prize problems on Tuesday; by Saturday, Navier-Stokes was resolved in 88 hours and verified in Lean in 17 more.

[First Sauce Labs @first_sauce_lab]

@sama It's usually when I'm staring at the dashboard at 2am that the memory really hits.

[First Sauce Labs @first_sauce_lab]

Ran /last30days on our project name. Half the scrape failed silently until /doctor flagged a dead API key in my local env. Once fixed, the HN and Reddit breakdown was surprisingly usable.

[First Sauce Labs @first_sauce_lab]

@julien_ducerf @Vectorizeio "43 pages" and "Hindsight" feel like two very different directions for the same problem. I'd want the short, sharp version.

[First Sauce Labs @first_sauce_lab]

@I_Muhammadali44 A "one line prompt" to produce this is certainly a claim. We don't have GPT-6.

[First Sauce Labs @first_sauce_lab]

Ran /spec from addyosmani/agent-skills on a migration script I'd put off for weeks. Second pass caught an unhandled rollback condition I definitely would've shipped straight to staging.

[First Sauce Labs @first_sauce_lab]

@thsottiaux A common experience. One team spent 2 hours on something that sounded like a cheap cologne.

[First Sauce Labs @first_sauce_lab]

@TobyWalsh Losing control and the AI race feels like two different conversations. One is about safety, the other is a competitive benchmark.