
October 2026 is already one of the busiest AI months of the year. This week's standout releases are practical tools: an API that answers in probabilities, two breakthroughs in AI voice, a record-cheap language model, and an assistant built for one-person businesses. Here is what each one does, what it costs, and who should try it.
1. OpenAI Decisions API: Answers in Probabilities, Not Paragraphs
Released in public beta on October 6, the Decisions API is a new endpoint (POST /v1/decisions) built for one job: fast classification. Send it text or an image with typed questions, and instead of generating text it returns a verdict. Three question types are supported: a predicate (a 0-to-1 probability that a statement is true), a choice (one option from your list, with a confidence score), and a score (a rating against ordered levels you define).
It runs only on OpenAI's gpt-6-luna model at $0.10 per million input tokens, with nothing charged for output or caching. OpenAI says it is about 10 times faster than the regular Responses API, making it ideal for support-ticket routing, content triage, and agent workflows needing thousands of quick verdicts a day.
2. Suno Speech: Voiceovers and Music in a Single Take

Suno, known for turning text prompts into songs, has moved beyond music. Its Speech feature, in public beta since October 1, generates a spoken voice reading your script while composing original background music to match. Voice and score come from one model as a single cohesive track, so there is no mixing to do yourself.
Simple mode writes the words from a one-line description; Advanced mode performs your custom script with controls for voice gender, style, and variation. Clips run up to about eight minutes, and a toggle mutes the music for clean narration. It uses Suno's existing credits, and beta quirks like drifting accents mean it suits intros and meditations better than polished commercial work for now.
3. ElevenLabs Eleven v4: Direct Your Voice Like a Film Director
ElevenLabs launched Eleven v4 and its low-latency twin v4 Turbo on September 28. The headline feature is scripted expressiveness: write direction tags straight into your text, and the model performs them instead of reading them aloud. Tags can be stacked, and context stitching keeps delivery consistent across long passages.
The model covers more than 90 languages (up from about 70), handles up to 10,000 characters per request (roughly ten minutes of audio), and clones a voice from just 10 seconds of audio. v4 Turbo targets voice agents at ~100ms latency, and API pricing starts at $40 per million characters. Best for: audiobooks, dubbing, narration, and voice agents.
4. Claude Haiku 5.5: A Tiny Model With a Million-Token Memory

Anthropic released Claude Haiku 5.5 on October 7, calling it the cheapest, fastest, and most capable small model it has ever shipped: $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, which Anthropic estimates is around 75% cheaper to run than Haiku 4.5 on average.
Cheap does not mean weak. Haiku 5.5 carries a 1-million-token context window with up to 128K output tokens, adaptive thinking, and an adjustable effort setting, scoring 72.4% on the OSWorld computer-use benchmark versus 15.7% for its predecessor. It is available on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry. Best for: summaries, classification, and high-volume subagent tasks.
5. IrisGo for Solopreneurs: The AI That Watches How You Work
IrisGo, the startup backed by Andrew Ng's AI Fund, has launched a version for solopreneurs and small business owners who spend hours a week on recurring admin. Its "Watch and Learn" approach flips the usual chatbot model: show it how you do a task once, and it automates it from then on, running on your device. It ships with a skills library for email drafting, invoice processing, report building, and summarization, and your data stays on your machine.
The beta is free and runs on Windows 11 and Apple Silicon Macs. For freelancers juggling inbox, meetings, and expenses, the pitch is simple: hand off the repetitive workflows and spend the day on customers.
Quick Comparison: Which One Is For You?

| Tool | What it does | Price | Best for |
|---|---|---|---|
| OpenAI Decisions API | Typed verdicts from text or images | $0.10 per 1M input tokens | Classification and routing systems |
| Suno Speech | Voice plus original music, one track | Beta; uses Suno credits | Narrated intros, meditations |
| ElevenLabs Eleven v4 | Expressive TTS with tags, 90+ languages | From $40 per 1M chars | Audiobooks, dubbing, voice agents |
| Claude Haiku 5.5 | Fast LLM, 1M-token context | $0.10 input / $0.50 output per 1M | High-volume dev work |
| IrisGo | Learns and automates admin workflows | Free during beta | Freelancers, one-person businesses |
If you build software, start with the Decisions API or Haiku 5.5. If you create content, Eleven v4 is the professional pick while Suno Speech is the fun experiment. And if repetitive admin eats your workday, IrisGo's free beta is worth a try.
Disclaimer: Prices, features, and availability above are based on October 2026 announcements and third-party reporting, and are subject to change. Always check the official product pages for current pricing and terms before purchasing.