RussianCow

Terra without reasoning is almost certainly going to be faster than Luna on High (which is roughly equivalent in intelligence according to AA).

That is very much not what "pump and dump" means.

My guess is nothing, and they're currently just releasing them individually as soon they're ready instead of all at once so that they hit the market more quickly. AI moves quickly enough that a delay of a couple weeks could result in significant lost revenue.

That margin just needs to be lower with LLMs than it is with only humans. In a lot of fields, that's a low bar. For example, medical professionals make a shocking amount of mistakes in aggregate.

> thinking real hard about what to say the next time they get together

LLMs can help tremendously with this part, though. Exploration/brainstorming, discovery, data analysis, etc can be made significantly easier with LLMs in most knowledge based fields.

This is my experience as well. Sol and Astra are by far the best of the models I use at following the instructions in AGENTS.md, almost to an annoying point. (But I'd rather have that problem than the inverse.)

This is clever and I can't believe I didn't think of it. I'm so used to appending "don't make any changes" to the end of my prompts.

The older Kimi K models (2.5 and 2.6) weren't bad for the time, but all of their models have had problems with overthinking. This also makes them significantly more expensive in practice than their pricing suggests, since they produce way more output (reasoning) tokens than most other models. Anecdotally, I've also had problems with all the pre-K3 models thinking themselves in circles trying to solve very simple problems that other models didn't struggle with. (I haven't played with K3 much because it's just too expensive.)

Extrapolating a bit, Moonshot AI provides Kimi K3 at $15 / $3 / $0.3 per million output/input/cached tokens. A 30% reduction would put K3.1 at roughly $12 / $2.4 / $0.24.

For comparison:

    Sol 6.1:    $10 / $2  / $0.10
    Opus 5.5:   $20 / $4  / $0.20
    Astra 6:    $50 / $10 / $1
    Fable 5.1:  $50 / $10 / $0.25

So Kimi K3.1 Pro would be generally competitive with the latest Sol and Opus models on API pricing. And that's before factoring in third-party providers, some of which serve Kimi K3 for significantly less than Moonshot. For example, inference.net offers K3 for $10.80 / $0.239 / $0.21.

(One caveat: A Claude or GPT subscription still gives you way more value compared to API pricing.)

Exciting times in any case! We really need more direct competition to OpenAI and Anthropic.

---

Edit: As another comment pointed out, official pricing is now available and it's about 2/3 the price of Kimi K3: https://platform.kimi.ai/docs/pricing/chat#model-pricing

Borrow checker aside, I would like to think that its type system inherently makes it safer than a language like C.

I 100% agree and am a big fan of DS V4.1 Flash in general for this reason, but I'll add my own caveat: Sol 6.1 isn't far behind on completion time for tasks because it's so damn efficient. I regularly get ~300 tokens/sec with DeepSeek Flash, but it spends the majority of that time thinking and executing slow commands one by one, whereas Sol thinks less and is much smarter about the commands it runs, resulting in far fewer turns to get the same thing done. For highly agentic tasks, DS might only be ~25-50% faster, not 500% as the TPS may suggest.

(But to be clear: 25-50% faster is still a big win in my book.)

But we're not comparing a Claude subscription to using Claude at API prices; we're comparing it to using a cheap model like DeepSeek V4.1 Flash at API prices. If I hit the 5-hour limit in 2 hours with Claude and have to wait 3 hours for it to reset, I get less value out of that subscription than I would have if I'd have just used DeepSeek to begin with, and the latter isn't more expensive.

So my original point about the limits being awkward still stands.

I tried it but it's way too slow for me. DeepSeek V4.1 Flash is an order of magnitude faster, and the modest difference in intelligence isn't worth it to me for most use cases.

This is my experience with all MiniMax models: They're good on paper but fail to live up to expectations in practice. IMO, at no point has any MiniMax model been the superior choice over others, all things considered.

I don't understand what you're saying. If I hit my 5-hour limit in 2 hours, I still need to get work done for the other 3. I'm not going to sit there and wait for my limit to reset.

> Claude sub has a 60x subsidy, so a $20 sub gives you $1200 in API.

This is only true if you max out the plan, which requires working around awkward 5-hour and weekly limits. For side projects, sure, you can use a Ralph loop or whatever and just continue when the window resets. For work, I can't afford to simply stop working when I hit my limit. And using Claude at API prices is prohibitively expensive.

Personally, I have a $20/month OpenAI subscription that I supplement with DeepSeek V4.1 Flash (currently using CoralBricks as the provider since it's fast and I'm grandfathered into their $0 cache read pricing). It's working out rather well so far. Even when I use DS all day as my primary model, I only pay about $1/day, which is roughly equivalent to what I pay for the sub anyway (but with no limits).

> It's an especially strange metric with open-weights decision models. I mean I can run those on my laptop ... cost is very abstract there.

Almost no one is running DeepSeek V4.1 Flash on their laptop, so comparing based on API pricing seems perfectly reasonable to me.

> As an aside, it's really unfortunate that artificial analysis only provides comparison at the highest reasoning level, which is mostly unusable for high-volume production use cases. I would really like them to start providing comparisons at low and medium reasoning levels across these models because those tend to be the ones that actually get used in production.

The frustrating thing is they started doing this for the big labs—you can see their benchmarks for all reasoning levels of Sol 6.1 or Haiku 5.5. They just don't seem to do it for any of the open weight models.

You don't need to directly compare the same reasoning levels. Today you can compare Sol on Low vs Haiku on Max, for example, but you can't look at GLM 5.3 High, only Max.

I hate to break it to you, but this is true of returns across a lot of companies and product categories. If you return unopened wine to Costco, for example, they throw it out because they can't guarantee that you stored it properly.

> Almost every test between just Opus and Opus conducting cheaper agent swarms ends up with Opus getting better results at a lower cost.

Do you have any links to these tests? I've had a hunch that this was true for a while but haven't found any benchmarks showing it.

I think it also helps that speed and latency aren't as vital for coding and we can afford to let the model think for longer.

Yes, but that's why you pin them. If you specify more than one provider in `order`, OpenRouter will use the first one unless it's down, so that's the only time it would switch providers on you. And personally, I'd pay a few cents instead of waiting for the API to come back.

> one only has to look at the providers like OpenRouter and OpenCode to see the massive swing since this summer

Doesn't this have more to do with LLMs getting more usage overall? I don't see how that's a sign that people are leaving OpenAI/Anthropic.

Just pin your config to a single provider, or several providers with the params `order` and `allow_fallbacks: false`. I regularly get ~98-99% cache hit rates with OpenCode. And some providers are much faster than DeepSeek; I was getting 200-300 tokens/second the other day with Together as my provider.

It's regrettable that OpenRouter doesn't even try to pin you to a single provider per session, but once you know about it, it's a problem that's easily solved.

It's fast. The average speed is 115 tokens/sec according to OpenRouter. I haven't tested the model to see how it is in practice, but I'd certainly pay a little extra for faster inference.

Edit: Though the average latency of 1.5s isn't very low, so it might not be that fast in practice for agentic work. Also, I don't know how much thinking it does, as that's generally been the drawback to Chinese models.

Is it? I would guess that it's much more about the race to get customers as they both near IPO than anything to do with the Chinese models.

I said "most companies considering paying Anthropic", which is not the same as "most companies". I also don't agree that "nobody" is doing this; I have lots of anecdata suggesting otherwise. Maybe the majority of companies are using the easy option of Copilot or Gemini like you said, but it's nowhere near 99%.

It's not that weird. Most companies considering paying Anthropic are probably not considering Chinese models as alternatives. Many don't even realize they exist.

I said this in another comment, but Artificial Analysis has the cost per task of Haiku on max roughly equal to that of Sol on medium, and the latter is significantly more intelligent. (And I'd wager that Sol probably finishes tasks more quickly, even with Haiku inference being faster.) So Haiku really only makes sense on lower reasoning levels, and only if you care about intelligence and speed more than you do about cost effectiveness (where Luna currently dominates). And that's without even bringing Chinese models into the mix.