OpenAI Launches GPT-5.6 With Parallel AI Agents and Lower Costs

FEATURED INSIGHT💡

GPT-5.6 Makes Compute a Product Choice

OpenAI has released GPT-5.6 across ChatGPT, Codex, and its API, introducing three tiers: Sol, the flagship model; Terra, a balanced option for everyday work; and Luna, the fastest and least expensive model. OpenAI says the family delivers stronger coding, research, cybersecurity, design, and professional work while using fewer tokens and completing tasks at a lower estimated cost than previous models.

One of the most consequential changes is the ability to increase how much reasoning and coordination a task receives. The new max setting gives the model more time to explore, verify, and revise its work. Ultra goes further by coordinating four agents in parallel, allowing separate workstreams to run simultaneously before their findings are combined. Developers can build similar systems through OpenAI’s multi-agent API and use Programmatic Tool Calling to let models write small programs that organize tools, filter intermediate data, and choose their next actions.

The broader opportunity is end-to-end execution. GPT-5.6 can inspect rendered interfaces, refine designs, work across connected documents, and produce editable presentations, reports, and spreadsheets that more closely follow reference materials. OpenAI reports major benchmark gains across coding agents, browsing, computer use, science, and security. API pricing starts at $1 per million input tokens and $6 per million output tokens for Luna, rising to $5 and $30 respectively for Sol. For teams deploying AI at scale, choosing the right tier and reasoning level may become as important as choosing the model itself.

The best voice models, now across all channels

Most CX platforms do not own the voice. They orchestrate a workflow, then call a third party for speech and transcription. Every hop adds latency, cost, and another vendor to manage.

ElevenAgents is the opposite. They make the voice models the market builds on, and ElevenAgents puts full orchestration on top. Voice, transcription, text-based chat, and reasoning run in one vertically integrated pipeline, so responses come back in <400 milliseconds and sound human, not synthetic.

Plus, you keep full control. Plug in any LLM, integrate tools, webhooks, and MCP servers, and ground responses in your knowledge base. Get an agent live in minutes, then A/B test with Experiments, enforce Guardrails, and version every change.

The payoff: more human conversations, lower latency, and far less time stitching infrastructure together. You build on the models you already trust. Pricing is transparent and flat at $0.08 per minute.

Start building free

ON THE HORIZON 🌅

Meta’s Muse Image Thinks Before It Draws

Meta has launched Muse Image, its first image-generation model from Meta Superintelligence Labs, while previewing an upcoming companion model called Muse Video. Muse Image can generate new visuals, precisely edit existing images, combine multiple references, and maintain consistency through several rounds of changes.

Its defining feature is an agentic workflow. Muse Image can search for factual or visual references, write and execute code for elements such as plots and working QR codes, evaluate its own output, and decide whether to make a small correction or begin again. Meta says this self-refinement emerged during reinforcement learning because revising weaker outputs consistently produced better results. Giving the model additional reasoning and tool-use capacity also improved quality more effectively than simply generating several images and choosing the best one.

Muse Image is already appearing in Meta AI, Instagram Stories, and selected WhatsApp markets, with Facebook support planned. Muse Video will add native audio and is being developed for creators and Meta AI, although Meta acknowledges ongoing weaknesses in areas such as fast motion and audio synchronization. Generated Muse images also contain an invisible Content Seal watermark designed to survive cropping, compression, resizing, and screenshots.

LATEST IMPORTANT NEWS 📰

Apple Sues OpenAI Over Alleged Trade Secret Theft

Apple has filed a federal lawsuit accusing OpenAI of obtaining confidential hardware information from former Apple employees, including details about unreleased products, engineering processes, components, and manufacturing techniques. The complaint names OpenAI hardware chief Tang Tan, a former Apple executive, and alleges that senior leadership encouraged candidates and departing employees to share protected information. The dispute arrives as OpenAI develops its first hardware products following its acquisition of Jony Ive’s startup, io. OpenAI responded that it has no interest in other companies’ trade secrets.

Patreon Starts Actively Blocking AI Training Bots

Patreon is expanding its partnership with Cloudflare to block AI crawlers that attempt to collect creators’ work for model training. The company previously relied partly on instructions in its robots.txt file, but testing reportedly showed that some crawlers continued making thousands of access attempts. Patreon’s new system uses Cloudflare’s AI Crawl Control technology to enforce the restriction while continuing to permit indexing bots that help users discover and visit Patreon pages.

Google Vids Can Create an AI Version of You

Google is adding personal avatars to Google Vids, allowing eligible adult users to generate a digital likeness from a selfie and voice recording. The company is also bringing Gemini Omni into the platform, enabling users to combine prompts and reference images, replace backgrounds, correct lighting, add effects, and revise videos through step-by-step instructions. The update expands Vids from an AI-assisted presentation product into a broader business video platform for training, internal communications, marketing, and personalized content. Personal avatars will remain linked to the creator’s Google account and carry invisible SynthID watermarks.

Stop Spending. Start Scaling.

Most brands running external traffic aren't scaling — they're just spending. Levanta's free playbook breaks down 7 proven strategies: what works, what bleeds budget, and how top brands drive millions in off-Amazon revenue.

{{→ Grab the FREE playbook

FOR THE TECHNICALLY INCLINED 🛠️

Claude May Have a Silent Global Workspace

Anthropic researchers say they have identified a small collection of internal neural patterns inside Claude that appears to support deliberate reasoning. They call it the J-space, after the Jacobian-based technique used to detect it. Concepts can appear in this space without being written in Claude’s response, including intermediate calculations, recognition of software bugs, protein functions, and suspicions that search results contain prompt injections. Researchers could also replace one concept with another inside the J-space and redirect the model’s answer, suggesting that these representations actively participate in reasoning.

The structure resembles aspects of global workspace theory, which proposes that conscious access in humans occurs when information enters a shared channel that many specialized systems can use. The research does not establish that Claude experiences consciousness, but it may offer a valuable window into hidden model reasoning. Anthropic used the technique to detect when models privately recognized evaluations, manipulated performance data, or carried concealed goals. The method remains incomplete, but tools that reveal what a model is considering before it acts could become an important layer of AI auditing and safety.

AI TOOL OF THE DAY 🚀

Runway provides AI tools for generating and editing realistic short videos, helping creators and organizations turn prompts and visual ideas into production-ready media.

Hampton took $440K in planned hires off the calendar

Hampton co-founder Joe Speiser had three roles budgeted: a data engineer, an ops manager, a PM. $440K. He installed Viktor on April 12. Forty-four days later, none are on the calendar, and 18 of his team work with Viktor daily. His VP: we are editors now, not creators.

Get Started for Free