Good Morning! Here’s what I have for you in today’s newsletter:
Nous Research launches Bot Mode for Hermes Agent
Topview brings Seedance 2.5 to 1080p
OpenAI previews Ultrafast, running GPT-5.6 Sol up to 14x faster
How to stop overpaying for Claude Code in 2026
4 new AI tools worth trying today
AI AGENTS
Nous Research introduced Bot Mode for Hermes Desktop, an alternative to sessions mode where each agent profile becomes a persistent "bot" with its own job, description, and profile picture, able to communicate with other bots, now open for a one-day public beta through a plugin.
Bot Mode replaces the sessions-based structure with one ongoing chat per agent profile, so a bot retains its identity and purpose across every interaction instead of starting fresh each session.
Bots can be given specific jobs and descriptions, then communicate directly with other bots, opening the door to multiple specialized agents coordinating instead of one general-purpose assistant handling everything.
The one-day public beta runs through a dedicated plugin, with Nous Research explicitly asking for feedback on bugs and behavior before folding it into the main Desktop app for everyone.
Giving an agent a persistent identity rather than a disposable session changes how it can be used day to day, closer to assigning a task to a specific teammate than starting a new conversation each time. Bots communicating with each other points toward multi-agent workflows becoming a native feature rather than something developers have to build themselves.
AI TOOLS
Topview brought Seedance 2.5 to 1080p resolution, delivering more cinematic output and richer detail for anyone generating video directly on the platform, positioning it as one of the most production-ready AI video models currently available.
The jump to 1080p specifically targets output quality, giving creators sharper, more detailed footage than the resolutions Seedance 2.5 previously shipped at on the platform.
Topview frames the update around cinematic quality rather than raw generation speed, suggesting the target use case is finished, presentable video rather than quick drafts.
The update is live now directly inside Topview, meaning existing users get the resolution bump without any separate setup or new account required.
Resolution has been one of the clearer, most visible gaps between AI-generated video and footage people would actually use in a finished project. A move to 1080p closes part of that gap directly, making Seedance 2.5 output usable in more contexts without an upscaling step afterward. For anyone producing content on a platform like Topview, this is a concrete quality upgrade rather than a new feature to learn.
AI MODEL
OpenAI opened a limited preview of Ultrafast, a new API service tier that runs GPT-5.6 Sol up to 14 times faster than Standard processing, generating up to 750 output tokens per second through a partnership with Cerebras while preserving the model's full intelligence and quality.
Ultrafast runs the exact same GPT-5.6 Sol model and weights as Standard processing, meaning the speed gain comes purely from Cerebras's wafer-scale chip architecture rather than any change to the model's capability.
The preview is currently limited to a select group of API customers testing use cases across coding, financial research, customer support, and voice applications, with no published pricing yet.
OpenAI is using the rollout period to study where an order-of-magnitude speed increase creates the most real-world value before expanding access more broadly.
Speed and intelligence have historically been a tradeoff, forcing a choice between a smaller, faster model and a slower, more capable one. Ultrafast breaks that link entirely, delivering frontier-level intelligence at speeds previously only available from lighter models. For latency-sensitive work like live voice interactions or real-time incident response, this changes what's actually possible to build.
HOW TO AI
Here's what's actually driving the number, and the habits that keep it low.
You've probably watched two Claude Code sessions do the exact same job and cost completely different amounts, one wrapped up in five minutes, the other burned through twenty, and you never figured out why.
Being efficient here doesn't mean using fewer tokens overall. It means making sure the ones you spend go toward the thing you actually asked for. Two questions decide that. What a single token costs. And how many of them a session ends up sending.
What actually decides the price of a token
You're billed per token, but what you're paying for is inference, the GPU time it takes to run the model over your request. Output runs about 5x the price of input, because generating a response happens one token at a time. Reading is cheap. Writing is expensive. Then there's prompt caching: if a request starts with exactly the same tokens as one the server just saw, it reuses that state instead of recomputing it, at a tenth of normal input price.
Run /clear between tasks
Nothing in a session gets sent once. Every file Claude reads and every command it runs gets appended to the conversation and resent, cheaply through cache, on every turn after it for the rest of that session. /clear when you're starting something genuinely new. /compact when the earlier part of the same task is done but you're staying in the same lane.
@-mention files instead of naming them
Type a file path and Claude may need to go find it first, a search, then a Read call. @-mention it and the file gets attached directly to your message, so that search step gets skipped. You only need to mention it once per conversation, it stays in context after that.
Going further: automate the habits
ccusage reads your local usage logs into daily, weekly, monthly, and per-session cost reports, plus a live view of your 5-hour billing window, the plain-numbers tool for seeing exactly where the spend is going.
P.S. You can access all the AI trainings, prompts and workflows if you upgrade.
GitHub shipped Balanced code review depth for Copilot, now generally available, giving developers deeper analysis and more thorough pull request reviews compared to Lite, with the default configurable at the org, repo, or individual PR level.
Notion added Custom Agent triggers that run automatically after an AI Meeting Note finishes, letting teams update project trackers, share recaps in Slack or email, and turn feedback into engineering tickets without manual follow-up.
Z.ai discovered a potentially serious vulnerability in Cursor using GLM-5.3 on a complex reverse-engineering task, disclosing it privately to the Cursor team, who are now working closely with Z.ai on a fix before public details are shared.
🎬 Runway: Nano Banana 2 Lite now on paid plans, warp-speed image creation without compromising quality.
🗄️ Polygres: turns your entire Postgres database into a context window for AI agents, relational, graph, and vector search in one hybrid query.
🎵 Mubert API: generate and stream royalty-free AI music for apps, games, and AI products, text-to-music and image-to-music built in.
🎥 Higgsfield: free Seedance 2.5 1080p generations for all new users, high-budget camera work and native 10-bit color, limited time.
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial