[cogman10]

Unlike all your examples, switching out an LLM is both cheap an easy. So easy that every 3 months or so new models are released and people grab them and start using them.

The UX is the same regardless the provider. You send in a prompt, it spits back an answer.

In all your other cases, the cost to switch is losing support and a difficult transition period. But in the case of LLMs, there was no support to begin with. The transition is basically updating your current harnesses to know about the other models.

I think the comparison most apt is the rise of AMD. Sure, it never(?) achieved market dominance, but it did ultimately make a huge dent. And a big part of that was because AMD x86 was pretty close and pretty compatible with Intel x86 at a fraction of the cost.

[pants2]

If you're developing on top of LLM APIs directly, this is definitely not true. There are differences in how context caching works, in what's available through native harnesses, the types of tools you're fine-tuned on (GPT uses apply_patch while Claude uses edit, with different formats), the API surface (Agents SDK, Responses API, Managed Agents), cost structures, and best-practice guidance all around.

Not to mention the meta of account limits, billing, ZDR contracts, etc.

[peab]

It really depends on what you're doing, but most LLM usage and agentic runs are pretty interchangeable in my experience, and it's usually trivial to switch.

If anything, you're better off supporting multiple LLMs as backup because most model providers have been so inconsistent with working all the time

[saberience]

Dude it’s not trivial to switch because the behaviors are different!

You’re clearly not building a product based on an LLM.

I’m still using various old Anthropic and OpenAI models for products I’ve built and released because I can’t risk the behavior changing in unpredictable ways and the users being pissed.

It’s much easier to switch out some deterministic software than an LLM which you’ve spent a ton of time on testing and benchmarking and understanding its nuances. Changing it is like replacing an employee who’s critical to the business.

[staticman2]

Anthropic has discontinued models in as little as 13 months from launch so if you do business with them switching can't be that big a deal?

[vrganj]

Can you give specific examples on what differences make it hard to switch?

Because this claim is counter to my experience as well.

[mjhay]

The behavior of a single model and version can and does change. There’s not only built-in stochasticity, but closed hosted models like Claude are tweaked and changed all the time.

[moxza]

For the public facing consumer functionality I have Gemini Flash running on guardrails directed by a state machine that calls it statelessly everytime. For that, it's strictly locked to a version. I can't afford to suddenly get responses that the SM is not tuned for.

As for which model does the building... I'm not at all attached. Enough logic, and CI gates/tests live outside the whims of the LLM to be able to hotswap them any time.

[asdfaoeu]

I don't think they are saying it's trivial but compare say for example switching an organisation from Office or Windows the example that started this. They are not even in the same ballpark.

[peab]

I've been building multiple products with LLMs, and they are in fact interchangeable for the most part.

In fact, most benchmarks show this! Most benchmarks have similar performance for the same classes of models.

On top of this, there are tools like open router, or even the openai SDK which trivially allows you to swap endpoints for the LLM!

If you're using the agents SDK from openai or something, then yeah it's not interchangeable but that's you doing it wrong

[devsda]

Makes sense but honestly if you've spent more time testing and working around the nuances to build consistent experience doesn't it mean you actually need more standardization to easily switch models if/when your trusted model is not viable for you provider?

[byzantinegene]

just use your agents to do the migration, that's what it's good at.

[chairmansteve]

Exactly, the LLM itself can do the migration, surely?

[tharkun__]

Exactly, as in, really, will they? Where and at what price, especially across an actual enterprise that needs to deploy them to lots of devs? There's much more than just the actual model.

Of course my numbers are a sample of one and I am not spending a lot of money or time on it. Just lazily trying things on my "happen to have this" hardware. But basically trying out the Claude Code I'm used to from work but locally with a bunch of open weight models.

I can run super tiny models on my 8GB NVIDIA card. They all suck (I have to use <=~5GB models if I want "usable" ~250k context that doesn't need to use system RAM and CPU (which makes things super slow).

I've also tried a GLM 4.7-flash, which even though it's super slow (in comparison) with ~250k context and it just doesn't cut it vs. the Claude Sonnet or Opus I get to use at work. All the while these are all touted as "totally usable, Claude/ChatGPT killer!" replacements.

It's just not "there" with tool use or building software for that matter. Like, just a simple Claude "web search" fails with it. So I asked it to build itself its own "web search" functionality and it just couldn't. It made so many mistakes its just not funny any more. And it couldn't recover from them either. I retried a few times (as I didn't have python installed and it wanted to implement it using that - this happens to be new system - never mind other attempts). I spent as much time doing this (and failing) as I spent building an actual full feature at work last week w/ Sonnet.

If it can't build itself a simple web search to .md file tool/skill, how am I supposed to trust this with actual coding? I'm used to being able to point Claude at our large code base and essentially work with it like a junior doing my bidding. Maybe 5.2 is a killer game changer vs. what I was able to try out (if slowly) but you really have to show me to convince me at this point. And not with synthetic benchmarks. In those, all of the models I tried are supposedly super awesome.

[arcanemachiner]

4.7 Flash is a small model that's almost a year old, which is ancient. And yes, your dinky GPU will not run anything worthwhile.

Just spend $5 on OpenCode Go and give GLM 5.2 a shot if you have the time. It's not quite as good as Opus, but it's more than good enough for many tasks.

[mikae1]

> Just spend $5 on OpenCode Go

$5 the first month, then price is doubled.

[arcanemachiner]

The $5 is so they can see if open weights models are worth using, not so they can use it for a month. (Which you can't; The quota runs out way sooner than a month for any serious usage. Still worth the price of entry.)

[miroljub]

If you use DeepSeek v4 Flash as a daily driver, with an occasional usage of DeepSeek V4 Pro and Glam 5.2 when necessary, the monthly quota practically never runs out.

[Sevii]

With OpenRouter you can pay flat rate and try nearly any model on the market.

[wonnage]

Open weights != local models.

[Der_Einzige]

This is a meme and massively over-complicating what is ultimately quite simple.

[AussieWog93]

>switching out an LLM is both cheap an easy.

Honestly, these days probably less friction switching out Redis or Elasticsearch (backend) than changing LLM provider (human facing).

Fable is seriously good enough now to, in a 20k line project, take "replace Mongoengine with raw PyMongo" and not screw anything up.

[woeirua]

Agents will make all of these migrations trivial. I expect margin collapse across a lot of tech darlings.

[thedougd]

This is the conversation I plan to have with Okta sales soon. Wait till you see how easy AI makes it to switch to Entra ID or anyone else. It’s tedium not even problem solving.

[calgoo]

My problem with the SSO providers is not the technical part, thats "easy". Its the coordinate with the 200+ external and internal vendors / support to redeploy the SSO part which is time consuming. I always say its a ~3 year project, which can be done in 6 months with the right amount of resources, especially if the platform has been running for years.

[thedougd]

There are the handful that it’s not just a web console or API you interact with. Have to file a support ticket.

Those will be a pain.

[itzprintz]

If the price is the only incentive, I'd stay away from Entra ID

[vidarh]

If the point was to move there'd be no point in having a conversation with Okta sales. The point is to get a discount.

[jaxn]

though Okta is the first provider working on the enterprise mcp stuff.

[thedougd]

It’s a new SKU. You’ll have to pay for that new type of app.

[mewpmewp2]

How could switching out a stateful service be easier than stateless?

[calebhwin]

Hard disagree.

Two LLMs with the same numbers on important benchmarks could have vastly different behavior in actual deployment. Not sure if as hard to switch as Excel <> Libre but still not "cheap and easy".

[vmg12]

This is just another example of the bitter lesson. In a year a model will come out that will make none of these model specific optimizations you made matter.

[pizlonator]

Yeah

But the point is that at any moment, there is friction in switching

[onion2k]

Unlike all your examples, switching out an LLM is both cheap an easy.

Rolling out AI access in a large business is still hard, especially if you're trying to do it safely e.g stopping people throwing all your company data including user PII into a chat for productivity reasons.

It's more a staff training and guardrails issue than a choosing which LLM to use issue, but I imagine picking an open model like GLM would make it harder because the 'enterprise stuff' will be missing.

[andsoitis]

> So easy that every 3 months or so new models are released and people grab them and start using them

Individuals perhaps, but not organizations.

[iknowstuff]

I don’t exactly see orgs lining up to switch (and train) their employees between claude desktop and codex and whatever copilot is doing. There’s probably some inertia to those harnesses/integrations on top of the llms themselves.

[Escapade5160]

Most large orgs do not need to train end users. They just need to add glm-5.2 to their router and their in house harness will pick it up. Then slowly limit usage on anthropic models and people will swap willingly. It's a simple /model command in every harness.

[torginus]

Yeah, most big orgs are pushing the idea of 'whitelabel' LLMs. Even if they choose to hang on to Claude Opus, they won't name it, they'll just call it the 'extra mode' and 'auto mode' will eventually switch to a local LLM in their harness.

[mgambati]

The inertia is legal and financial. People are paying Anthropic through AWS accounts because the simple reason of not dealing making new contract and legal agreements is enough of reason of the inertia.

But, eventually, I’m quite sure that AWS will also provide open models with those contracts without any inertia. Copilot is already offering Kimi.

My company has a deal with Devin and they provide new models all the time, and open models are becoming the most used ones by our internal metrics, especially because the company is very worried about cost.

[zmgsabst]

AWS already supports Llama and GLM in its Bedrock service for hosted models.

They’re much cheaper to run, eg, Llama 3.3 Instruct 70B is 5-10x cheaper than Sonnet 5.

https://aws.amazon.com/bedrock/pricing/

Say you have 20% of usecases that require the more expensive model — but in 80% you could just use Llama instead of Sonnet (eg, for basic queries of a document). That saves 80% of that 80%, or 65% of your total bill!

That is the kind of “swap” that’s likely to occur in automated tooling as pricing pressure kicks in — “can you save 65% on our AI bill by switching Bedrock over in 80% of uses?”

[regularfry]

Bedrock is really out of date with the models it offers, to the extent that I'm not sure they even have plans to update what's on there now they have the deal with Anthropic. They're still offering Qwen 3, not even 3.5 and certainly not 3.6. GLM 5 is the newest z.AI model they have, when it's 5.2 that would be the one to worry Sonnet.

There are some ok models on there (Qwen 3 Coder Next is usable and fast, for instance) but the lack of updates in a fast-moving field makes it something I don't want to recommend to my org.

[whattheheckheck]

Because these models are not going to stand up legally

[regularfry]

That's a stretch but it doesn't change the outcome.

[skeptic_ai]

Also they pay for legal liability of code produced

[Yizahi]

Maybe for a fantasy of legal liability of output produced. I haven't heard of any LLM corpo being held liable for any output they generate. Even NYT lawsuit is going nowhere for 3 years in courts already, despite being the most grounded.

[skeptic_ai]

Copilot insures your company up to x if someone sues our company. I guess most companies do this. That’s why we can’t use anything else except copilot.

[bayarearefugee]

What "training" do you have to do to get a professional developer to switch LLMs or harnesses? Its literally just download the other one, point it to your code base and start typing into that text box instead of the other one.

[iknowstuff]

devs are the easiest, yeah.

[saghm]

What would "training" even entail for that? As far as I can tell, using these tools directly is basically identical in terms of what you need to know. If you happen to have a bunch of custom configurations, maybe you need to invest some time into porting them, but it's not clear to me why you think that anyone would need to be trained if they spent months using one tool and then suddenly had t switch to the other.

[peab]

Enterprises switched from openai to anthropic this year - anthropic overtook openai for the first time. I don't see why they wouldn't switch again.

There's barely any moat. All the data is with connectors, memory is near useless

[bag_boy]

Switching an agent harness is more difficult, especially on the enterprise/teams level.

Once your team gets settled with Claude teams, cowork, and the various plugins, it’s going to be a pain in the butt to switch.

[walthamstow]

Plugins and skills are completely trivial to move and most work with any model. What is not trivial are Anthropic's new managed agents vendor lock-in offering.

[danny_codes]

Is it? I switched to Kiro and it's essentially identical.. well a bit better because you get a better idea of what the harness is doing, but otherwise identical.

[bag_boy]

I should have clarified that it’s not difficult for technical workers to switch, but for your average knowledge worker…it’s not easy switching from Claude Cowork to Codex. I don’t think the switch from Cowork to open source is easy.

[Shorel]

I made my own. It's true I don't want to switch to another harness now.

But switching models is just a command.

[skissane]

The irony is that Claude will help you migrate away from itself

AI is possibly the first product in history that will eagerly help you replace it with one of its competitors.

[skeptic_ai]

Wait till they will pop a warning: “using Claude to migrate away will suspend your account. “

Or even better just silently sabotage the migration so you can’t do it. Something we can definitively expect from Claude given past behavior

[Forgeties79]

> Unlike all your examples, switching out an LLM is both cheap an easy.

For now

[rhipitr]

Switching out an LLM? What do you mean by this? Sure some models can run locally but in a company with lots do people they might not be willing to spend to self host a larger model that requires beefier hardware to host, plus all the complexity to scale that out to a bit internal user-base

[KetoManx64]

Most of the AI companies have OpenAi compatible API's, so you just get a subscription from another provider and change the URL that your LLM Agent Harness uses to talk to the AI.

I use OpenRoutet which lets you switch between providers (Anthropic, ChatGPT, Z-AI) whenever you want. Sometimes I'll have two different models from different providers evaluate each other's answers.