| Claude jumps a tier, ChatGPT gets a memory that dreams, DeepSeek undercuts everyone, and three conferences worth your summer. |
| Model Production |
SAMPLE ISSUE
Sat · June 13, 2026 |
Your weekly read on every AI assistant that matters — Claude, ChatGPT, Gemini, Copilot, Perplexity, DeepSeek, Grok, Mistral & more.
|
|
Welcome to the sample issue. One newsletter, every major AI assistant — so you can skim the week in five minutes instead of chasing fifteen changelogs. This week was a big one: Anthropic shipped a model a whole tier above what was public just a few days ago. Let's get into it.
|
|
★ Headline Story
Anthropic brings “Mythos” to the masses with Claude Fable 5
On June 9, Anthropic released Claude Fable 5 — its first publicly available Mythos-class model, a full tier above Opus 4.8 and previously locked to government cybersecurity partners. In plain terms: the most capable Claude ever is now something you can actually use.
The benchmarks are the story. On SWE-bench Pro (hard engineering tasks) Anthropic reports 80.3%, versus 58.6% for GPT-5.5. One early customer says Fable 5 finished a frontier physics research task in 36 hours using a third of the reasoning tokens GPT-5.5 needed over four days. It also completed Pokémon FireRed from raw screenshots alone — no maps, no game state — something earlier Claudes couldn't do without scaffolding.
The catch is price. API access runs $10 / $50 per million input/output tokens — roughly 2× Opus 4.8, but less than half what the Mythos preview cost. Batch pricing drops it to $5 / $25.
What “Fable” vs “Mythos” actually means: they're the same underlying model. Mythos 5 ships with no safety classifiers on cyber/bio queries and is restricted to approved partners; Fable 5 layers classifiers on top, routing flagged requests to Opus 4.8 instead.
|
Why it matters: The frontier just moved, and it moved into general availability. Fable 5 is live now on the Claude API, Amazon Bedrock and GitHub Copilot, and is free on Pro, Max, Team and Enterprise plans through June 22 — after which it shifts to usage credits. If you've been on the fence about a paid plan, the next two weeks are the cheapest this model will ever be.
|
|
|
|
⚡ The Wire — this week, by assistant
|
| ChatGPT |
OpenAI gave memory a brain. Plus and Pro users get 2× memory capacity, plus a new “dreaming” behavior that quietly refreshes saved memories over time so they stop going stale. Also new: Lockdown Mode (opt-in protection against prompt-injection data exfiltration), an Active Sessions security panel, and answers that render as interactive charts inline. Heads up — GPT-4.5 retires June 27. |
|
|
| Gemini |
Fresh off I/O, Google's app picked up a Daily Brief, a redesigned “Neural Expressive” interface, the Gemini Omni video model and a personal agent called Gemini Spark. On the plumbing side, Gemini in Chrome for Android (with Nano Banana + auto-browse) lands late June, and Gemini 3.5 Flash is now on by default across Enterprise. |
|
|
| Copilot |
Build 2026 was Copilot's show. Microsoft unveiled the Copilot super app and Scout, an always-on agent that acts without being re-prompted, plus seven new in-house MAI models (including the 35B-parameter MAI-Thinking-1). Browsing with Copilot brings agentic, multi-step web tasks to the enterprise, and Excel gets “Edit with Copilot.” |
|
|
| Perplexity |
Perplexity wants to be your work layer, not just a search box. Its Comet browser went free worldwide, “Computer” expanded into Word, Excel, PowerPoint and Outlook, and Max subscribers now get Claude Opus 4.8 access. At Computex it showed a hybrid local-cloud router that decides mid-task what runs on your PC vs the cloud. ARR reportedly crossed $450M. |
|
|
| DeepSeek |
The price disruptor is still the one to watch. The V4 preview (1.6T-param Pro, 284B Flash, 1M-token context) holds its specs through June, with Flash at just $0.14 / $0.28 per million tokens — near-frontier quality at a fraction of everyone else's cost. General availability is the next domino. |
|
|
| Grok |
xAI shipped fast this week. Grok Voice rolled out publicly (June 4) alongside Grok Imagine 1.5, a new image-to-video model with natural-language motion control and a Quality Mode for sharper generations. Musk confirmed a core model upgrade on June 5, with Grok V9-Medium (1.5T params) finishing training and a public release expected mid-June. A dedicated coding model, Grok Build 0.1, is in API beta. |
|
|
| Mistral (Vibe) |
Europe's frontier lab renamed Le Chat to “Vibe” — now one agent and one licence across work and code. New Work Mode handles long-range agentic tasks, Code Mode does remote coding and pull requests (with a VS Code extension), and Mistral Medium 3.5 (128B params) landed. Workflows orchestration is in public preview. |
|
| Also on our radar: Meta AI, Alibaba's Qwen, Amazon's Nova, and open-weight upstarts. We rotate the lineup each week based on who actually shipped — tell us which one you want covered. |
|
|
|
📅 On the Calendar
Where the AI world is gathering next.
|
Jul 6–12
Seoul |
ICML 2026 — One of the three premier ML research conferences, at COEX, Seoul. In-person. |
Jul 8–9
Paris |
RAISE Summit 2026 — Europe's biggest AI gathering: 9,000+ attendees, 450 speakers, heavily C-level. In-person. |
Aug 4–6
Las Vegas |
Ai4 2026 — America's largest AI conference: 12,000+ attendees, 1,000+ speakers at The Venetian. In-person. |
|
|
That's the week. If a friend forwarded this, you can subscribe at the link below — and reply to tell me which assistant you actually use day to day. It shapes what I cover next.
— The Model Production team
|
|
Subscribe & read past issues →
|
|
|
