This seems functionally similar to OpenAI having a step in pricing once you exceed a certain context length (also at 272k aka 2^18 aka 256k).
Having a lot of active context increases the per-token cost (flops issued and bytes read per token out) so it makes sense to pass that cost on to users. I'm actually surprised it's implemented as a hard cutoff instead of a smooth gradient.
For me the sweet spot is somewhere under 500k depending on how extensive I want to get. You can build up a sizable effort project in half a million tokens with Claude, with Claude having all the context from ground 0 to wherever you're off at.
I'm always curious what you guys are working on; every git repo I've run a local model on and stick below <100k to increase speed seems effective enough to scope patches and changes.
My current Claude Code session has been going on for like 35 hours and has used up around 400 million tokens, thankfully almost all of those being cached (95-98%) - pretty typical for long form agentic work.
First you spend like 2-3 hours working on a plan, once you have that you just tell the model to go and implement it, do adversarial sub-agent review loops before each commit and also make sure that all tooling and tests pass (including coverage requirements). You do need to poke it in a slightly different direction every few hours, though. Not even any novel work, just some refactoring and SSE notification hardening, bug fixes, alongside environment tuning and getting rid of some bottlenecks (also migrated from Oracle to PostgreSQL but that's mostly done).
That said, Kimi somehow manages to use less context in the main thread than Anthropic's models (even when you use sub-agents and also dynamic workflows in Claude Code), might have something to do with either how the model is tuned or their Kimi Code harness - because even in most of the longer form sessions it doesn't seem to fill up quite as quickly (note: because the kimi vis tool doesn't have a full summary view across all agents, these are the main long running agent stats across some sessions, not sub-agents):
I could see 256k context being sufficient for all sorts of work, even if intermediate progress/plan tracking files and docs might have to be used along the way, in addition to whatever plan support the harness has (for example, if you document something that will be relevant for load testing you might need that in 10 turns but not during the ones before then).
Once you have the plan you don't need to keep the 2hrs of research in the context (which is most of it) you can drop that plan into a file and start fresh for implementation.
Thanks for sharing - is this a normal feature request you are implementing in this example or is this a project from scratch? Trying to get an idea of how your workflow compares to mine.
I had Claude build me a Python-inspired .NET language that treats .NET as a first class citizen, and breaks backwards compatibility where some Python nuances don't really apply to .NET for. I was able to get it to build a sample ASP .NET Web application that ran on Culebral code.
Haven't gone back to it, have been using Claude Code on a private project I'm still architecting.
Try doing a refactoring of some sort or larger new feature using just an agent on a moderately sized codebase, 256k will be compacting every few minutes, and result will be unusable.
They are just talking to the model in CC, while staying in a single thread. Doubt they have any actual coding knowledge to compartmentalize different problems in the codebase.
Depends on the programming language I'm using for a given project, and the domain I'm working with. I've been coding as a hobbyist for nearly two decades now (since my teens), professionally for 9 years, and was a TA before that for roughly 3 years at one of the best colleges for this field in the state (at least back then it was) where I taught other students about programming, in some cases I was their primary learning resource.
But yeah, I have no idea about anything about software because you made an assumption off very little to go by.
Since Claude is the first time for me really, really out (TIL against my wished about https://status.claude.com/), I am now interested enough to see what else works. But ... when I click pricing, I see "Join a waitlist". Wtf? Are they really that good, so were totally surprised and overwhelmed by the requests, is this a marketing stunt, or do they just don't have the hardware being in china?
As a user, I much prefer that to service disruptions or severely degraded or secretly quantized performance. However if I didn't have an account, I'd be pretty pissed off about not being able to give them money and become a user.
The only downside with third party providers is that you have to trust that the provider have setup and configured it correctly, and is not secretly quantizing it.
Kimi had become that popular. I was a subscriber of Kimi back when latest version was Kimi K2. Later I unsubscribed because I jumped over to GLM subscription (they had amazing deal). Now when I wanted to try out Kimi K3 to find out what the fuzz was all about, I couldn’t subscribe to them.
I remember reading a post from Moonshot team about this, they are doing this because they are almost at peak capacity and want to reserve it to keep the quality for their current customers.
We are actually witnessing an open-weight model catching up at catching mainstream users attention. And instead of behaving like Anthropic, they actually care about their users experience.
It's an extremely popular model hosted by a company affected by hardware export bans. I doubt they'd voluntarily prevent people from subscribing if they didn't absolutely have to to maintain service quality.
It seems to only be available in Kimi Code, via subscription, no there's no API pricing. The linked page says it consumes about half as much quota as the 1M version though.
A bit of topic. But how likely is it that the US will restrict Chinese open weight models and also force Euro countries to do the same? I think it will be effective within 6 months. The US is having a hard time staying competitive.
No it's not. The US dictates every single step in Europe. Euro politicians are in the pockets of American institutions. The biggest reason of dumb decisions in the EU is because of deliberate decisions made in favour of the US. LNG, war in Ukraine, ASML export restrictions and many more.
Why are Anthropic and OpenAI even allowing their coding harness apps to be plugged into different model providers…? I’m surprised they haven’t figured out a way to clamp down on that by now.
From my view, as soon as they do that, they send people out the door to use Opencode instead - and once many people have a taste of trying every model via Openrouter, it's eye opening as to the possibilities.
Of course - Anthropic and OpenAI have an advantage in the amount they can subsidize the usage, but I think those days are waning.
Well when a huge part of potential revenue is all in on Bedrock... you need the harness to be able to talk to Bedrock. And Vertex. And all the other places these models are hosted. And allow for proxy because many businesses do not all direct internet access... all valid business reasons.
This was posted 38 minutes ago, and as of 20 minutes ago, several Anthropic services are now designated as having a "major outage".
Doubt these are related, but it made me laugh a little.
Anthropic services have outages on all days ending in y.
So users in Germany are not affected?
Wow. So kimi is suddenly half the price for all users until they hit 256k of context? Thats massive.
This seems functionally similar to OpenAI having a step in pricing once you exceed a certain context length (also at 272k aka 2^18 aka 256k).
Having a lot of active context increases the per-token cost (flops issued and bytes read per token out) so it makes sense to pass that cost on to users. I'm actually surprised it's implemented as a hard cutoff instead of a smooth gradient.
Not surprising it's a hard cutoff: they almost certainly have two infrastructure configurations for the two max sequence lengths
Fewer nodes dedicated to prefill per instance, and fewer nodes in total since you don't need to support a higher KV cache.
Disaggregated inference also means they can tune the balance of compute dedicated to prefill seperately from decode
That's actually nice! I usually try to stay below 200k context anyway.
For me the sweet spot is somewhere under 500k depending on how extensive I want to get. You can build up a sizable effort project in half a million tokens with Claude, with Claude having all the context from ground 0 to wherever you're off at.
I'm always curious what you guys are working on; every git repo I've run a local model on and stick below <100k to increase speed seems effective enough to scope patches and changes.
My current Claude Code session has been going on for like 35 hours and has used up around 400 million tokens, thankfully almost all of those being cached (95-98%) - pretty typical for long form agentic work.
First you spend like 2-3 hours working on a plan, once you have that you just tell the model to go and implement it, do adversarial sub-agent review loops before each commit and also make sure that all tooling and tests pass (including coverage requirements). You do need to poke it in a slightly different direction every few hours, though. Not even any novel work, just some refactoring and SSE notification hardening, bug fixes, alongside environment tuning and getting rid of some bottlenecks (also migrated from Oracle to PostgreSQL but that's mostly done).
That said, Kimi somehow manages to use less context in the main thread than Anthropic's models (even when you use sub-agents and also dynamic workflows in Claude Code), might have something to do with either how the model is tuned or their Kimi Code harness - because even in most of the longer form sessions it doesn't seem to fill up quite as quickly (note: because the kimi vis tool doesn't have a full summary view across all agents, these are the main long running agent stats across some sessions, not sub-agents):
I could see 256k context being sufficient for all sorts of work, even if intermediate progress/plan tracking files and docs might have to be used along the way, in addition to whatever plan support the harness has (for example, if you document something that will be relevant for load testing you might need that in 10 turns but not during the ones before then).Once you have the plan you don't need to keep the 2hrs of research in the context (which is most of it) you can drop that plan into a file and start fresh for implementation.
Thanks for sharing - is this a normal feature request you are implementing in this example or is this a project from scratch? Trying to get an idea of how your workflow compares to mine.
I had Claude build me a Python-inspired .NET language that treats .NET as a first class citizen, and breaks backwards compatibility where some Python nuances don't really apply to .NET for. I was able to get it to build a sample ASP .NET Web application that ran on Culebral code.
Haven't gone back to it, have been using Claude Code on a private project I'm still architecting.
https://github.com/Giancarlos/Culebral
I have a couple projects where the background research is easily over 500k without writing any code, after ultracode subagents synthesis.
Try doing a refactoring of some sort or larger new feature using just an agent on a moderately sized codebase, 256k will be compacting every few minutes, and result will be unusable.
They are just talking to the model in CC, while staying in a single thread. Doubt they have any actual coding knowledge to compartmentalize different problems in the codebase.
Depends on the programming language I'm using for a given project, and the domain I'm working with. I've been coding as a hobbyist for nearly two decades now (since my teens), professionally for 9 years, and was a TA before that for roughly 3 years at one of the best colleges for this field in the state (at least back then it was) where I taught other students about programming, in some cases I was their primary learning resource.
But yeah, I have no idea about anything about software because you made an assumption off very little to go by.
> k3-256k is now available. Within 256k context, it delivers the same results. k3 (1M) consumes about twice as much quota as k3-256k.
This is just an API level change right? The model itself should be the same I think.
This isn't quantized, right? Just a smaller context?
Its 256k context window. Quantization is orthogonal. We cant really tell directly so it could be quantized.
This is a fantastic option for swival.dev given its very efficient context management compared to e.g. Claude Code.
Since Claude is the first time for me really, really out (TIL against my wished about https://status.claude.com/), I am now interested enough to see what else works. But ... when I click pricing, I see "Join a waitlist". Wtf? Are they really that good, so were totally surprised and overwhelmed by the requests, is this a marketing stunt, or do they just don't have the hardware being in china?
> Are they really that good,
They did exceed the expecations of pretty much everyone! I've also blogged about using the model, it's a bit on the slow side but pretty good!
> so were totally surprised and overwhelmed by the requests ... or do they just don't have the hardware being in china?
Yes, this is mostly the case: https://x.com/Kimi_Moonshot/status/2078855608565207130
As a user, I much prefer that to service disruptions or severely degraded or secretly quantized performance. However if I didn't have an account, I'd be pretty pissed off about not being able to give them money and become a user.
"I'd be pretty pissed off about not being able to give them money and become a user."
I am now rather pretty pissed towards antrophic for stopping my flow and forcing me to search for alternatives.
Since the model is open-weights, you can get from other providers, for example see https://openrouter.ai/moonshotai/kimi-k3#providers
The only downside with third party providers is that you have to trust that the provider have setup and configured it correctly, and is not secretly quantizing it.
See Kimi Vendor Verifier.
No, the waiting list is true.
Kimi had become that popular. I was a subscriber of Kimi back when latest version was Kimi K2. Later I unsubscribed because I jumped over to GLM subscription (they had amazing deal). Now when I wanted to try out Kimi K3 to find out what the fuzz was all about, I couldn’t subscribe to them.
I remember reading a post from Moonshot team about this, they are doing this because they are almost at peak capacity and want to reserve it to keep the quality for their current customers.
We are actually witnessing an open-weight model catching up at catching mainstream users attention. And instead of behaving like Anthropic, they actually care about their users experience.
It's an extremely popular model hosted by a company affected by hardware export bans. I doubt they'd voluntarily prevent people from subscribing if they didn't absolutely have to to maintain service quality.
What is the purpose of this? Just a hard cutoff below the actual context window? You could set that in your harness anyway.
Uses less quota (i.e., cheaper). For people who like to keep their contexts small, this is a no-brainer.
> k3 (1M) consumes about twice as much quota as k3-256k
Cheaper?
k3-256 consumes half the quota, so... yes? (assuming the user isn't making use of the larger context window)
I can't seem to find pricing for this model. Since the context size is just a quarter of the full size K3, is the price also much cheaper?
I usually keep my context in chats below 256k anyways so this would be tremendous honestly.
It seems to only be available in Kimi Code, via subscription, no there's no API pricing. The linked page says it consumes about half as much quota as the 1M version though.
A bit of topic. But how likely is it that the US will restrict Chinese open weight models and also force Euro countries to do the same? I think it will be effective within 6 months. The US is having a hard time staying competitive.
I don’t know, but I do think that the days of the US “forcing” Euro countries to do anything, is over.
No it's not. The US dictates every single step in Europe. Euro politicians are in the pockets of American institutions. The biggest reason of dumb decisions in the EU is because of deliberate decisions made in favour of the US. LNG, war in Ukraine, ASML export restrictions and many more.
Trump burned a lot of bridges in the EU, that one will be a hard sell
omg! new model!!
Same model, new configuration
The bells tolled today, but nobody came to church.
Why are Anthropic and OpenAI even allowing their coding harness apps to be plugged into different model providers…? I’m surprised they haven’t figured out a way to clamp down on that by now.
I'd guess because it costs them nothing and it gives you a smoother transition back towards paying for their products.
From my view, as soon as they do that, they send people out the door to use Opencode instead - and once many people have a taste of trying every model via Openrouter, it's eye opening as to the possibilities.
Of course - Anthropic and OpenAI have an advantage in the amount they can subsidize the usage, but I think those days are waning.
Well when a huge part of potential revenue is all in on Bedrock... you need the harness to be able to talk to Bedrock. And Vertex. And all the other places these models are hosted. And allow for proxy because many businesses do not all direct internet access... all valid business reasons.