TL;DR On September 10, 2026, OpenAI announced three products: the Agents API (public beta) that exposes the same agent harness behind Codex, GPT-Live-1 voice in the API at $0.05/minute, and ChatGPT for Financial Services built on GPT-6 Astra. The first two matter most for everyday builders and small teams.

The short version

OpenAI's September 10 release day is less about a new model and more about packaging: it's taking the infrastructure that already powers Codex and ChatGPT and letting developers rent it directly. The Agents API gives you long-running, multi-agent workflows in a single API call, while GPT-Live-1 brings ChatGPT's natural full-duplex voice to any app[1][2].

Robot hand representing AI agents

1. The Agents API (public beta)

The Agents API lets developers build and run cloud agents with the Codex harness, fully managed by OpenAI[1]. You describe the task, model, tools, and environment, and OpenAI handles the rest — context management, tool use, and subagent coordination[1].

Three capabilities stand out. First, automatic context compaction keeps agents working across long sessions by compressing earlier context as a session approaches its limit[1]. Second, tool search and programmatic tool calling let agents load only the tool definitions they need and run calls in parallel — supporting MCP, custom functions, and built-in web search[1]. Third, multi-agent support splits complex tasks into subagents that work in parallel and then recombine results[1].

You choose where the agent runs: in an OpenAI-hosted sandbox, on your own infrastructure, or through sandbox partners including Cloudflare, DigitalOcean, Modal, Oracle, Runloop, and Vercel[1]. The harness itself is open source (the public Codex codebase), and there are no additional fees — you pay only for the tokens and tools your agents use[1].

2. GPT-Live-1 voice in the API

GPT-Live-1 brings ChatGPT's full-duplex voice — the model can listen and speak at the same time — to the API, with more control over how voice agents behave[2]. It's priced at $0.05 per minute for the front-end voice layer, paired with whatever backend model you choose[2].

  • Natural interruptions. A single model reasons over incoming and outgoing audio together, avoiding the latency of chained speech-to-text, LLM, and text-to-speech systems[2].
  • Delegation. GPT-Live-1 can hand deeper reasoning and tool calls to a backend model like GPT-6 Astra[2].
  • Telephony. It supports phone-call deployments, from restaurant reservations to customer support[2].
  • Developer control. You shape tone, pace, and style through the system prompt, and get native transcripts plus turn detection[2].

OpenAI reports GPT-Live-1 improves Full Duplex Bench by 30 percentage points over GPT-Realtime-2.1, and that language-learning app Speak saw interruptions during thinking pauses fall by almost 80% versus prior turn-based systems[2].

3. ChatGPT for Financial Services

For finance teams, OpenAI also launched ChatGPT for Financial Services — a ChatGPT Work experience that pairs GPT-6 Astra with built-in financial data from providers like Daloopa, PitchBook, LSEG News, and Crunchbase[3]. It's shaped by design partnerships with Morgan Stanley and Evercore, targets investment banking and equity research, and adds granular citations so analysts can trace figures back to source tables and passages[3]. Availability is limited to eligible financial institutions[3].

What it means for small teams

  • Agents without the plumbing. The Agents API removes the need to build your own orchestration, context compaction, and subagent management — the parts most small teams can't afford to maintain[1].
  • Voice support is now cheap to pilot. At $0.05/minute for the voice layer, a phone-based agent is a realistic experiment rather than a six-figure build[2].
  • Finance is enterprise-only for now. Unless you're an eligible financial institution, ChatGPT for Financial Services isn't something you can sign up for today[3].

The caveats

The Agents API is in public beta, so expect iteration and change as OpenAI works toward general availability[1]. GPT-Live-1 is a voice front-end, not a standalone assistant — its usefulness depends on the backend model you pair it with[2]. And while the Agents API has no platform fee, your token and tool costs still scale with usage, so budget before you fan out hundreds of agents[1].

Bottom line

Verdict: If you build software or automation, the Agents API is the most consequential release here — it commoditizes agent infrastructure that previously required real engineering. GPT-Live-1 is the sleeper pick for anyone who wants natural voice support without the cost of building it from scratch.

Frequently asked questions

Does the Agents API cost extra?

No platform fee. You pay only for the tokens and tools your agents use[1].

What does GPT-Live-1 cost?

$0.05 per minute for the front-end voice layer, plus the cost of the backend model you pair it with[2].

Can I use ChatGPT for Financial Services?

It's available to eligible financial institutions; you'd contact OpenAI's financial services team to get access[3].

Sources

  1. OpenAI — "Introducing the Agents API" — openai.com
  2. OpenAI — "Build more natural voice experiences with GPT-Live-1 in the API" — openai.com
  3. OpenAI — "Introducing ChatGPT for Financial Services" — openai.com