AI NewsOctober 8, 2026

Decision models return a single choice or probability, like “yes,” “spam,” or a 1-10 score, in one step instead of generating a full written response, making them far faster and cheaper for simple classification tasks.
OpenAI launched its Decisions API last week, running on GPT-6 Luna, charging $0.10 per million input tokens with no separate charge for the output since none is generated as text. Separately, open-source toolkit Unsloth released a way to fine-tune any existing small model, like Qwen3.8 or Gemma 4, into this same decision-making format, and says it raised one small model’s accuracy from 20.7% to 74.3% on three decision benchmarks using just 4GB of GPU memory.
These tools suit high-volume, repetitive choices, like sorting support tickets or flagging spam, where a full written explanation would be wasted and the person doesn’t need to understand the model’s reasoning.
Key Capabilities:

• No output tokens: Returns a direct answer instead of generating explanatory text, cutting cost and latency.
• DIY option: Unsloth’s approach lets developers convert their own small model into a decision model, not just use OpenAI’s.
• Low hardware bar: Unsloth’s fine-tuning runs on just 4GB of GPU memory.

Playground turns a text prompt into a browser game, no coding required, while a deeper Unity-powered version called Spark is coming later this year for more complex projects.
Google and Unity launched Playground on October 7, available now to US users 18 and older at playground.google.com. Creators choose genre, perspective, and single- or multiplayer mode, then refine the game by chatting, with uploaded images converted into matching game assets. Generation runs on weekly token limits; Google One AI subscribers get a larger allowance than free accounts.
Games can be played and shared via link but can’t be sold or exported to desktop Unity at launch. Unity’s Spark, a closed beta coming later this year, is positioned as a tool for ongoing creation rather than one-shot game generation.
Key Capabilities:

• No coding needed: Describe a game in plain text; the AI builds and lets you refine it by chatting.
• Age-gated launch: Limited to US users 18 and older at launch.
• Not sellable yet: Games can be shared via link but not sold or exported to desktop Unity.

SynthID Detector, previously limited to journalists and researchers, is now open to anyone at synthid.com, checking files for an invisible watermark Google has applied to over 180 billion images and videos since 2023.
Google launched the public version on October 7, supporting uploads in common formats like JPG, MP4, and MP3, with a daily limit of roughly 10 checks per account. The detector also catches content from OpenAI, Nvidia, and Kakao, which all use the SynthID watermarking standard, with Apple support planned.
A clean result only means no SynthID watermark was found; it can’t confirm a file is real, catch AI content from models that don’t use SynthID, or detect open-weight models that skip watermarking entirely.
Key Capabilities:

• Multi-company coverage: Detects watermarks from Google, OpenAI, Nvidia, and Kakao, not just Google’s own tools.
• Free and global: Live now in English worldwide, with a daily per-account check limit.
• Known limits: Can’t flag content from AI tools that don’t use SynthID, including most open-weight models.

MAI-Code-1.1-Flash lets GitHub Copilot complete coding tasks without sending any data to Microsoft’s servers, but it needs serious hardware: Microsoft’s own recommendation exceeds what most laptops have.
Microsoft announced local inference for MAI-Code-1.1-Flash on October 7, a coding model with 137 billion total parameters that only “activates” 6.8 billion of them per request, similar to how a large team might assign just a few specialists to a given task rather than involving everyone. The on-device, compressed version shrinks to 53GB, an 80% size reduction from the full cloud version, while Microsoft says it keeps most of its coding ability intact.
Developers can let GitHub Copilot automatically decide between cloud and local processing, or force it to stay fully on-device for sensitive work. Full rollout across Copilot’s CLI, app, and VS Code arrives by month’s end.
Key Capabilities:

• Runs offline: Completes coding tasks without any cloud connection or data leaving the device.
• Hardware-hungry: Microsoft’s recommended setup exceeds what most consumer PCs currently have.
• Hybrid option: Copilot can auto-switch between cloud and local models, or be forced to stay local.

With GPT-6’s new Intelligent UI, a question about splitting a bill can return an actual working calculator instead of a paragraph explaining the math.
OpenAI rolled out GPT-6 with Intelligent UI on October 7, letting ChatGPT compose answers using charts, tappable buttons, forms, and small interactive tools it builds on the spot, choosing the format based on the question. It began with Plus, Pro, Business, and Enterprise users, running on GPT-6 Sol, with Free and Go tiers getting GPT-6 Luna starting October 8.
OpenAI says this is an early step, with more work needed on how the model chooses layouts, and that it’s now also able to start displaying an answer while it’s still “thinking” through the rest of the response.
Key Capabilities:

• Live tool creation: Can build a working calculator or checklist directly inside a chat response.
• Tiered rollout: Paid tiers got it first; Free and Go followed one day later.
• No extra cost: Intelligent UI has no separate usage quota from normal chat messages.

Claude Haiku 5.5 costs 90% less than its predecessor for most requests and beats OpenAI’s rival small model, GPT-6 Luna, on every benchmark Anthropic tested head-to-head.
Anthropic released Haiku 5.5 on October 7 at $0.10 per million input tokens and $0.50 per million output tokens for prompts under 100,000 tokens, matching GPT-6 Luna’s price exactly. It’s the first Haiku model with an adjustable effort setting, letting developers trade cost for intelligence, and its context window grows to 1 million tokens, up from 200,000.
It completes Anthropic’s Claude 5.5 lineup, following Opus 5.5 and Sonnet 5.5 earlier this fall. Anthropic also quietly cut Sonnet 5.5’s cache-read price in half, from $0.20 to $0.10 per million tokens, in the same announcement.
Key Capabilities:

• Price: $0.10/$0.50 per million tokens (input/output) for prompts under 100K, 90% below Haiku 4.5.
• Bigger context: 1 million tokens, up from Haiku 4.5’s 200,000.
• First with effort control: The first Haiku model letting developers balance cost against intelligence.