Daily AI Launch Radar: North Mini Code, Claude Fable 5, and ChatGPT Memory – June 10, 2026

June 10 launch snapshot: Cohere’s North Mini Code was the most immediately testable release in this edition: an Apache 2.0 agentic coding model with a 30-billion-parameter mixture-of-experts architecture and 3 billion active parameters. Anthropic’s Claude Fable 5 pushed the frontier-model story toward longer-running knowledge work, while OpenAI’s new ChatGPT memory system focused on keeping personal context current over time.

This is a historical Radar for releases announced through June 10, 2026. Availability has changed since then, so each entry below separates the launch-day facts from later developments.

1. North Mini Code

Cohere released North Mini Code on June 9 as its first agentic coding model. The model has 30 billion total parameters, activates 3 billion per token, supports a 256,000-token context window and up to 64,000 output tokens, and is licensed under Apache 2.0.

The deployment range is the useful part. Cohere offers the weights on Hugging Face, API access, and managed deployment through Model Vault. Cohere’s current documentation says trial and production API keys can use the model free until applicable rate limits are reached. Teams considering it for production should still price the infrastructure, evaluation work, security review, and agent harness rather than treating the model endpoint as the whole cost.

Cohere reports competitive coding and terminal-agent benchmark results for its size class, including a 33.4 Artificial Analysis Coding Index score and internal throughput comparisons with Devstral Small 2. Those are vendor-reported results. Kingy has not reproduced the benchmark runs or completed a hands-on repository test.

Kingy verdict: North Mini Code is the clearest operator-facing launch in this set. Its open weights, permissive license, compact active footprint, and multiple deployment paths make it worth a controlled evaluation for teams that want more control over coding-agent infrastructure.

2. Claude Fable 5

Anthropic announced Claude Fable 5 on June 9 for complex coding and professional work that can span many steps and longer time horizons. At launch, Anthropic positioned it for ambitious knowledge work rather than as a lightweight default model.

The access story changed shortly after this Radar window. Anthropic suspended Fable 5 and Mythos 5 on June 12 after new US export controls took effect, then restored global access beginning July 1 after those controls were lifted. Anthropic currently lists Fable 5 for Claude Pro, Max, Team, and Enterprise customers, with API and cloud-platform availability subject to the provider’s current rollout and account terms.

Kingy verdict: Fable 5 matters because it extends the frontier-model competition into sustained, multi-stage work. Buyers should evaluate it against their own task duration, review burden, latency, and total usage cost. The launch announcement alone does not establish that it will outperform a cheaper model on a specific workflow.

3. ChatGPT memory “dreaming”

OpenAI’s June 4 memory update introduced a more scalable system for synthesizing and refreshing personal context from conversation history. OpenAI calls the background process “dreaming.” It is designed to revise memories as circumstances change, surface a reviewable memory summary, and reduce stale or contradictory context.

The initial rollout covered Plus and Pro users in the United States, with additional countries and Free and Go users planned over the following weeks. OpenAI said serving improvements cut the compute required for the system by roughly five times. Users can review and edit the memory summary, turn memory off, or use Temporary Chat when they do not want a conversation to use or create memories.

Kingy verdict: This is a meaningful product update, but it raises a practical trust question alongside the convenience benefit. Users should inspect the memory summary and controls before relying on personalization for sensitive work. A memory system that updates itself can be more useful than static notes, but it also needs transparent correction and deletion paths.

What to test

• North Mini Code: run a representative repository task, record hardware or API cost, and compare the patch and test results with the model your team already uses.
• Claude Fable 5: test one long-running research or coding workflow with explicit checkpoints and human review rather than relying on headline benchmark claims.
• ChatGPT memory: review the generated memory summary, correct one outdated detail, and confirm the controls match your privacy requirements.

Sources and related coverage

• Cohere: Introducing North Mini Code
• Cohere documentation: North Mini Code specifications and access
• Anthropic: Claude Fable 5
• Anthropic: Redeploying Fable 5
• OpenAI: Better memory for a more helpful ChatGPT

Browse the AI Launch Tracker for canonical launch records and later status changes.