GitHub rolled Kimi K2.7 Code into Copilot this week — the first open-weight model available as a selectable option in the model picker. Hosted on Azure, billed at provider list pricing, available on Pro, Pro+, and Max plans. Straightforward changelog entry.
The HN thread hit 417 points and 185 comments. And almost nobody cared about Kimi.
What they cared about: the pricing shockwave ripping through AI coding tools, and the quiet exodus to local models.
The Real Story
The article itself is a routine GitHub changelog: Kimi K2.7 Code, an open-weight model, is now generally available in GitHub Copilot. This is the first open-weight model offered as a selectable option in the Copilot model picker. Microsoft hosts it on Azure. You pick it in VS Code 1.127.0. Boring.
The HN thread turned into something else entirely.
“Gotta say, I’ve lost all interest in cloud-based AI products”
The top comment from user c7b sets the tone for the entire thread:
“Too many cool features and workflows that I was once excited about that I can’t or don’t use anymore for a variety of reasons (price hikes, subjectively nerfed, disappeared altogether, replaced,…) for me to even remember. It’s tiring. I’ve set up a small rig, mostly settled on Qwen3.6 and I’m slowly adding features myself. It probably can’t compete with Claude. I don’t even know, I’ve stopped checking. All it takes is to realize that it doesn’t actually matter if the grass is greener somewhere else. Feels so good to know that it won’t change under my feet.”
This isn’t an edge case. It’s the dominant sentiment in the thread. Devs are tired of the rug-pull cycle — price hikes, nerfs, shutdowns (RIP Gemini CLI), and unpredictable model quality swings. The escape route is local.
cyanydeez puts it more bluntly:
“I don’t know how anyone could have witnessed the last 2 decades of American VC funded tech startups and tell themselves, ‘you know, this will be a reliable technology with no hidden problems.’ Even a sober technical evaluation is just two steps: 1. You’re proposing to build an app on a non-deterministic model. 2. That model is hosted behind a non-deterministic system (model alignment, guardrails, cost/token pricing). So you want to build your app and you think you’re going to keep up with both?”
The thread splits between those going fully local and those hunting arbitrage between cloud providers. Nobody trusts the status quo.
The Copilot Pricing Shock
The thread’s other pillar: Copilot’s June pricing change. Multiple commenters describe the same experience — hitting their monthly budget in days.
User Kon5ole:
“Unfortunately the June pricing change for Copilot forced me personally as well as my entire department at work to switch to Claude Code. With copilot we were hitting a few dollars of extra spend over the included credits in April and May, then in June we started chewing through the monthly budget every 2-3 days. Just a completely insane price hike from the customer’s perspective, I don’t know what MS were thinking there.”
nsbk echoes: “Last month we consumed all the subscription credits by the 7th day, and had to top the extra credits up every 2-3 days. Last month was definitely not cheaper than a CC subscription. It actually triggered a cost savings effort across the Engineering org.”
The old $10 Copilot that gave you all the frontier models? Gone. User MangoCoffee: “GPT-5.4 used to be x1 for yearly subscriber but now it cost 6x. I run out of premium requests for just couple prompts.”
phillipcarter (Anthropic employee) adds context that suggests this is structural, not temporary: “The widespread belief in industry right now is that the per-seat pricing (which Copilot bailed from first) is going to go away in the near-term.”
Follow the money: tokens cost what they cost. The subsidy era is ending.
The Local AI Migration is Real
Against this backdrop, the thread’s deepest vein is practical advice on running models locally. User c7b sparked a sprawling subthread on hardware — Mac Mini M4 configs, Strix Halo benchmarks, Qwen 3.6 vs Gemma 4, quantization strategies.
User SwellJoe:
“A 4-bit quantization of either Qwen 3.6 27b or Gemma 4 31b will run on a 32GB Mac with a decent-sized context. 64GB gets you the full ~256k context. The 4-bit QAT version of Gemma 4 has practically identical performance to the full size version or the 8-bit version. In short, right now, 64GB is all you need for the best models you can self-host on anything short of five-figure machines.”
User deadbabe drops the real take:
“People want to make it seem like you need to always use the latest and greatest frontier models to be taken seriously as a developer. You really don’t need them. After a certain point, bigger models give diminishing returns. If you can get 80% of the productivity gain with a free local model, use the local model. It will still be way faster than doing everything by hand, but you also don’t have to pay for tokens to a cloud provider and the tools won’t be ripped away from you on a whim. This is the new attitude enlightened people should adopt. Reject the arms race.”
The Harness Matters More Than the Model
A secondary thread that deserves attention: the importance of the agent harness versus the model itself. User irthomasthomas: “Claude in Claude Code has been shown to perform persistently worse in evals than Claude + a minimal harness.”
User taspeotis: “The harness is super important, what tools are available and the system prompts vary from harness to harness. Anthropic seems to have a modest lead on their harness and models, so it’s a best-of-both-worlds scenario.”
This cuts to the core of what Kimi K2.7 in Copilot actually means. The model is a commodity. The harness — Copilot’s system prompts, context management, tool integration — is where the value lives. Microsoft adding another open-weight model changes less than you’d think.
My Take
Two forces are pulling in opposite directions. Cloud AI is getting more expensive while local models (Qwen 3.6, Gemma 4) are getting shockingly capable on consumer hardware. The HN thread documents a migration in progress — not wholesale, but the early adopters who were first into cloud AI are now the first out.
The Kimi K2.7 launch is a sideshow. The real signal is that devs are pricing out the math and choosing ownership over access. When a base Mac Mini can run a 27B parameter model at usable speed, the calculus changes.
The arms race isn’t the open-weight models vs the frontier. It’s local sovereignty vs cloud convenience. And the gap is shrinking fast.
Article: Kimi K2.7 Code is generally available in GitHub Copilot (GitHub Changelog, July 1 2026) Discussion: Hacker News (417 points, 185 comments)
Further reading:
- Why I Stopped Using Copilot and Went Full Terminal Agent
- What 8 Coding Agents Cost Per Month
- State of Open Source Coding Agents 2026
Want to save money on AI coding tools? Check out best deals and discounts at aiFiesta.