Sponsored by
n
OpenAI just split GPT-5.6 into three models.
You now pick your fighter: Sol for speed, Terra for balance, or Luna for deep thinking.
This isn't an upgrade it's a complete rethink of how AI models work.
But wait, there's more:
Mistral built a robot brain that navigates with just one camera
Anthropic killed free access to Claude Fable 5 now you pay per token
I'm Alex. Welcome to L8R by Innov8.
Let's dive deep 🐰
In today's post:
OpenAI splits GPT-5.6 into three models: Luna, Terra, and Sol
Mistral's Robostral Navigate lets robots see with just one camera
Anthropic kills free Fable 5 access, forces users to pay per token
OpenAI Drops GPT-5.6: Meet Sol, Terra, and Luna
OpenAI just changed the game with its new GPT-5.6 model family.
Instead of one giant AI, you now get three different models built for speed, balance, or heavy thinking.
This new lineup finally gives developers real choices for building complex AI agents without going broke.
The details:
The flagship model, Sol, costs $5 per million input tokens and handles complex, long-running agent tasks.
Terra acts as the balanced middle ground, while Luna is the cheapest and fastest option at just $1 per million input tokens.
All three models boast a massive 1-million token context window and can output up to 128,000 tokens.
A new feature called Programmatic Tool Calling lets the AI write and run JavaScript right inside an isolated V8 engine.
Sol crushes the Agents' Last Exam benchmark with a score of 53.6, beating Claude Fable 5 in professional workflows.
Why it matters:
You no longer have to pay top dollar for simple AI tasks.
This three-tier system lets teams route easy jobs to Luna and save the expensive Sol model for hard coding or research.
It makes building real AI agents cheaper and way more efficient.
💡 L8R's Take:
OpenAI is done playing nice.
Calling out competitors' benchmarks as broken shows they want to own the agent space completely.
Sol looks like an absolute beast, but I bet Luna is where the real money gets made.
From Our Partner
See the whole platform. No guided tour.
Skip the sales call. Walk through Gladly's interface yourself — the AI suggestions, the unified customer view, the full conversation thread. 15 minutes, no installation, no commitment.
Mistral's New AI Lets Robots See With One Cheap Camera
Building a robot just got a lot cheaper.
Mistral AI just dropped Robostral Navigate, a new model that lets robots walk around using just one normal camera.
They do not need expensive lasers or radar to see where they are going anymore.
The details:
The 8 billion-parameter model takes simple text commands like "walk to the supply room" and figures out how to get there.
It drops the need for pricey LiDAR and depth sensors, relying completely on a single standard RGB camera feed.
It beats the best single-camera systems by 9.7 points, hitting a 76.6% success rate in brand-new environments.
You can put this brain into wheeled, legged, or flying robots because it works with almost any hardware.
Mistral released this on July 8, 2026, but it is locked behind enterprise deals right now instead of being open-source.
Why it matters:
Hardware is the hardest part of robotics.
Sensors like LiDAR cost a fortune and break easily.
By letting robots navigate with a cheap camera, Mistral tears down a massive wall for robot makers.
This pushes us closer to a world where smart robots are cheap enough to be everywhere.
💡 L8R's Take:
Mistral is playing a very smart game here.
While OpenAI and Google fight over chatbots, Mistral is quietly taking over the physical world.
I just wish they made it open-source for hobbyists instead of locking it up for big companies.
From Our Partners
The World's Biggest Dev Event Hits Silicon Valley
WeAreDevelopers World Congress comes to San José, CA — September 23–25, 2026, in the heart of Silicon Valley. 10,000+ developers, 500+ speakers, 20+ stages, and the full software development lifecycle in one place.
Kelsey Hightower. Thomas Dohmke (fmr. CEO, GitHub). Christine Yen (CEO, Honeycomb). Mathias Biilmann (CEO, Netlify). Olivier Pomel (CEO, Datadog). The people actually building the tools you use every day — all on one stage.
Three days of AI, agents, cloud, security, and architecture, plus workshops, live coding, and the official Congress party. Prices rise as the event gets closer — lock in today's rate.
Use code GITPUSH26 for 10% off.
Anthropic Kills the Flat Rate for Claude Fable 5
Anthropic is pulling the plug on 'all-you-can-eat' access for its smartest AI.
If you want to use Claude Fable 5, you will have to pay by the token starting July 12.
Flat-rate subscriptions just cannot keep up with the massive compute costs of these powerful models.
The details:
Fable 5 will cost $10 per million input tokens and $50 per million output tokens.
That price tag is exactly double the cost of the older Claude Opus 4.8 model.
The change removes Fable 5 from Pro, Max, Team, and Enterprise plans on July 12, 2026.
Anthropic claims this is a temporary move until they can secure more compute power.
Why it matters:
This is a massive wake-up call for developers and founders.
You can no longer throw the smartest model at every simple problem without burning cash.
You have to get smart about routing your data. Use cheaper models like Sonnet 5 for basic tasks and save Fable 5 for the heavy lifting.
💡 L8R's Take:
Flat-rate AI pricing is a dying dream for top-tier models.
Compute is simply too expensive right now.
Get used to paying for exactly what you use, because the rest of the industry will absolutely follow this trend.
🚀 Quick L8R Summary
GPT-5.6 Family: OpenAI ditched the one-size-fits-all approach and split GPT-5.6 into three models Sol, Terra, and Luna so you can pick based on speed, smarts, or cost.
Robostral Navigate: Mistral built an 8B model that lets robots navigate anywhere with just a single camera no LiDAR, no depth sensors, no fancy hardware.
Claude Fable 5 Paywall: Anthropic kicked Fable 5 out of subscriptions and moved it to pay-per-use billing at $10 per million input tokens and $50 per million output tokens..
📩 Innathe L8R engane undarunnu 👇?
We read every reply - just reply to this email and let us know how we can improve !
Appo adutha L8R il kanaam bie…👋