Over the past week, the AI community was captivated by a mystery. An unnamed model appeared on public testing platforms, and within days it topped usage leaderboards, reportedly processing tens of trillions of tokens as developers put it through its paces. Stripe CEO Patrick Collison called it "very impressive." This week, as reported by Bloomberg, the mystery was solved: the model is GLM 5.3 Flash, a new open-weight frontier model with weights released for anyone to build on. 

Here's why that matters for MSPs and their clients: GLM 5.3 Flash is already live in the Hatz AI platform. 

Intelligence is getting cheaper, fast 

The numbers tell the story. On leading benchmarks, GLM 5.3 Flash scores roughly 90 percent of the performance of Claude Opus, one of the most intelligent premium models available, at about one thirty-third of the price. It outscores mid-tier premium models that cost 15 times more. In practical terms for Hatz users, tasks routed to GLM 5.3 Flash use on average 33 times fewer credits. 

"We're getting to a point where intelligence is becoming more optimized for spend, and people's credits are going to stretch longer," said Alex, Tech Lead Manager at Hatz AI. 

And this is the pattern, not a one-time event. GLM 5.3 Flash joins a wave of open-weight models added to Hatz this year, including DeepSeek V4 Flash and Pro, GLM 5.2, Kimi K2.7 Code, and Nemotron 3 Ultra. Each scores within a few points of the frontier models at a fraction of the credits. The same week GLM 5.3 Flash arrived, OpenAI cut prices on its GPT 5.6 models, and Hatz passed those reductions straight through to the platform. 

The lock-in problem your clients don't see coming 

When a business commits to a single AI vendor, they lock in today's price for today's intelligence. When the market shifts, and it now shifts weekly, they are stuck renegotiating, re-architecting, or overpaying. 

Hatz takes the opposite approach. Your clients get secure access to more than 90 leading models in one platform, refreshed as new models ship. When a breakthrough like GLM 5.3 Flash lands, it shows up in their workspace within days, not quarters. And because Auto Model draws from the full catalog, their apps and workflows come along automatically. No more apps stuck on last year's model. 

Auto Model does the optimizing for them 

Most users should never have to read a benchmark chart. Hatz Auto Model Selection reads each message and sends it to the model that fits, picking per message, not per chat. It selects the right model 99.78 percent of the time, picks almost instantly, and replies start about 20 percent sooner than before. As conversations go on, Auto remembers what it has already read instead of paying for it again, cutting credits by up to 50 percent over the course of a chat. 

The savings are measured on Hatz, not modeled. In a real sales workflow benchmark, Auto Lite completed the entire job for about 490 credits per run, 23 times fewer than a pinned flagship model at 11,377 credits, and scored slightly higher on quality. Auto Performance and Turbo scored highest of all, at 8x and 6x fewer credits than the flagship. 

Every time a model like GLM 5.3 Flash arrives, Auto has a better option for everyday work. That is how it saves credits while raising quality at the same time. 

And for clients who want direct control, there are three modes to choose from: Lite for fast, cheap everyday work, Performance for the right model matched to each task, and Turbo for maximum brainpower. Admins can cap modes per role, set credit limits, and disable any model. Every model also remains directly selectable. Optimization by default, choice on demand. 

Access without the security tradeoff 

Going direct to a new model provider means new terms, new data handling questions, and a new risk review every time your client wants to try something. Within Hatz, GLM 5.3 Flash, Claude Opus, GPT 5.6, and every other model run inside your client's tenant on a SOC 2 Type II certified platform. Client data is never used to train models. Data stays in secured US data centers, separated by tenant. Open-weight models are hosted in the US with US-based inference providers, so nothing goes to a non-US lab's API. Model providers operate under zero data retention agreements, and every request is visible in audit and invocation logs. 

New intelligence, same protection, no new vendor risk assessment every time the market moves. 

How to message this to your clients 

Keep it to three points. 

• Hatz delivers the most intelligence at the most optimized cost. Frontier-level performance with a platform that continuously re-optimizes as prices fall. Their AI spend gets more valuable over time, not less. 

• No lock-in, full control. One model vendor means one bet. Hatz means 90+ leading models, with the freedom to choose or let Auto Model decide, and admin guardrails that keep spend inside the lines. 

• Enterprise security by default. Every model, including brand new open-weight releases, runs under SOC 2 Type II controls, US-only hosting, no-training guarantees, and zero data retention agreements. 

The AI market is moving faster than any procurement cycle. With Hatz, your clients don't have to keep up. The platform does it for them, and you get the credit. 

Not a partner yet? Sign up to experience complete control over your AI: security, spend, and scope.