Hacker News Reader

Comment by pyeri | Hacker News Reader

Have you tried one of the Kimi K2 models or the latest GLM models by z.ai? The general consensus is that they're at least at par with Claude's class.

View on HN

If K2 or GLM 5.2 are on par with Opus 4.8 I'll eat my hat. They're good, but they're not that good. Deepseek V4 Pro has been better than Sonnet for me, but the only model that comes close to or surpasses Opus 4.8 is GPT-5.5.

Honestly just give it time. This stuff moves so fast next month the conversation will be different. For folks who don’t like the ID privacy issues, use Deepseek et al and it should be able to get the job done even if the experience takes a bit more wrangling.

The problem with the ID verification is that they can pair introspective conversations with ID. Either that bothers people or it doesn’t.

Main point: we can’t fret about current state models because the ID verification has future implications. Models will change and competition will catch up. Do what feels right in the long run not whether TODAYS model is better at Anthropic.

I agree with this, my disagreement was strictly with saying that the current open models are as good as Opus.

They're not. And by the time they are Open AI and Anthropic will probably be onto the next thing.

Not sure what happened to Google in all this. They're falling out of the frontier race.

Both Anthropic and OpenAI don't want to continue training models indefinitely.

Anthropic CEO has expressed potentially slowing down on model training. There is little return for billions of dollars burnt for 1-2% increase on various benchmarks. These companies profit via inference.

Not to mention, the whole Fable being banned by the US Gov is a scary prospect for future models. What is the point of spending billions if its going to get blocked?

Of course this can't go on forever. Especially not on LLMs. But are we really close to the limits of what these LLMs can do? I'm not sure we are.

The difference between GPT-5/Opus 4 and GPT-5.5/Opus 4.8 is striking. For software development anyway, there's no comparison. And all this has happened in a year.

My assumption is there will be another 2-3 years of improvements ahead of us on LLMs alone. Through hardware upgrades, larger training runs, better data quality, better algorithms, etc.

Of course, by then these models will be quite expensive. Will my company pay for it? I don't know. I'm sure some people will though.

some people will, but they will have to bear the costs being the only users of llms sustained through billions of dollars of funding.

GLM 5.2 is far better than deepseek V4. Seriously feels like I’m talking to a Claude model. Also burns tokens like one, so there is that. Deepseek is unbeatable on price/quality.

They are but from our evals for example GLM 5.2 (unquantized) performs as well as Opus but uses more tokens and takes more time.

I really wish this would change soon but they are not there yet.

Using even double the total tokens and taking, what, 2-3x the time?, still seems worth it if prices are 5x+ cheaper (which OpenRouter [1] claims is the case).

On NeuralWatt for my personal projects at home (not affiliated, just a happy customer), I get so much more mileage out of GLM than I get out of Claude at work, specifically because it's priced as a hammer I can pound any nail-shaped-object with, not a delicacy I need to carefully budget-analyze to try to figure out if it's worth burning my monthly spend limits on this task.

https://openrouter.ai/compare/z-ai/glm-5.2/anthropic/claude-...

I thought true token use was being hidden by anthropic and openai both

No, they do specify token counts, as they let you pay for them. They just don't tell you what these thinking tokens actually are.

Though because they don't show you, they could be lying about it. Very unlikely, I think, would be too dangerous IMO. But technically possible

Reply on news.ycombinator.com