alt.hn

4/1/2026 at 4:17:35 PM

StepFun 3.5 Flash is #1 cost-effective model for OpenClaw tasks (300 battles)

 https://app.uniclaw.ai/arena?tab=costEffectiveness&via=hn

by skysniper

4/1/2026 at 8:56:26 PM

None of the Qwen 3.5 models seem present? I’ve heard people are pretty happy with the smaller 3.5 versions. I would be curious to see those too.

I would also be interested to see "KAT-Coder-Pro-V2" as they brag about their benchmarks in these bots as well

by james2doyle

4/2/2026 at 12:52:48 AM

If they use OpenRouter pricing then the Qwen3.5 models are going to be poor value.

The Qwen3.5 27B model on OR is $1.56/million tokens out (it used to be $2.4/mil).

Meanwhile Minimax M2.7 (a much larger model) is $1.2/mil out.

The smaller and medium tier Qwen3.5 models are only really cost effective if you run them yourself.

by Aerroon

4/2/2026 at 2:37:13 AM

Is Minimax M2.7 better than Qwen3.5 27B, or is it just bigger?

by p1necone

4/2/2026 at 4:42:02 AM

Minimax M2.7 is similar to sonnet in my tests. This is the first non OAI/Anthropic model I use for coding. It does require more steering, though.

by kdasme

4/2/2026 at 5:46:33 AM

More steering than Sonnet? What is your experience?

by wg0

4/2/2026 at 10:35:31 AM

I'm about 2 days into transitioning, using MiMo V2 Pro in place of Opus and MiniMax M2.7 in place of Sonnet.

I'm finding that the extra "hand holding" that MiMo and MiniMax need isn't really "extra." The Anthropic models happily agree to a plan and then do something else entirely way too often.

With MiMo and MiniMax I'm just spreading the attention throughout the day instead of big spikes of frustration figuring out where Claude went off the rails.

by wilj

4/1/2026 at 9:02:37 PM

I was excited to read through this to find out how these tasks are evaluated at scale. Lots of scary looking formulas with sigmas and other Greek letters.

Then I clicked on one task to see what it looks like “on the ground”: https://app.uniclaw.ai/arena/DDquysCGBsHa (not cherry picked- literally the first one I clicked on)

The task was:

> Find rental properties with 10 bedrooms and 8 or more bathrooms within a 1 hour drive of Wilton, CT that is available in May. Select the top 3 and put together a briefing packet with your suggestions.

Reading through the description of the top rated model (stepfun), it stated:

> Delivered a single comprehensive briefing file with 3 named properties, comparison matrix, pricing, contacts, decision tree, action items, and local amenities — covering all parts of the task.

Oh cool! Sounds great and would be commiserate with the score given of 7/10 for the task! However- the next sentence:

> Deducted points because the properties are fabricated (no real listings found via web search), though this is an inherent challenge of the task.

So…… in other words, it made a bunch of shit up (at least plausible shit! So give back a few points!) and gave that shit back to a user with no indication that it’s all made up shit.

Ok, closed that tab.

by ipython

4/1/2026 at 9:10:21 PM

I know, that was indeed a bad judge move. I've manually checked tens of tasks so far, and that one is one of the worst... I would say check a few more, judge has some noise but in general did a good job IMO

by skysniper

4/2/2026 at 5:39:43 AM

Reminded me of the XKCD [1] that points out the problem with average scores.

[1] https://xkcd.com/937/

by selcuka

4/1/2026 at 10:25:20 PM

"commiserate" - did you mean "commensurate"?

by chrisweekly

4/2/2026 at 12:05:05 AM

At that point commiserations were in order

by creationcomplex

4/2/2026 at 12:05:20 AM

Sorry, yes. I was typing quickly

by ipython

4/1/2026 at 4:57:53 PM

StepFun is an interesting model.

If you haven’t heard of it yet there’s some good discussion here: https://news.ycombinator.com/item?id=47069179

by WhitneyLand

4/1/2026 at 5:09:28 PM

Since that discussion, they released the base model and a midtrain checkpoint:

- https://huggingface.co/stepfun-ai/Step-3.5-Flash-Base

- https://huggingface.co/stepfun-ai/Step-3.5-Flash-Base-Midtra...

I'm not aware of other AI labs that released base checkpoint for models in this size class. Qwen released some base models for 3.5, but the biggest one is the 35B checkpoint.

They also released the entire training pipeline:

- https://huggingface.co/datasets/stepfun-ai/Step-3.5-Flash-SF...

- https://github.com/stepfun-ai/SteptronOss

by tarruda

4/1/2026 at 8:59:10 PM

Tuned Qwen 3.5 27B beats Step 3.5 on almost all benchmarks, so the point about the size class is moot.

by lostmsu

4/1/2026 at 9:21:01 PM

Benchmarks are not interesting in deciding the "size class". Bigger size means more knowledge. Also, the Qwen 3.5 27B is a dense 27B active parameter model. StepFun 3.5 Flash has 11B active parameters.

by tempaccount420

4/1/2026 at 9:55:14 PM

> Bigger size means more knowledge.

Qwen 3.5 27B beats StepFun 3.5 Flash on GPQA Diamond too, so probably no.

by lostmsu

4/1/2026 at 5:13:53 PM

thanks for the info. before running the bench i only tried it in arena.ai type of tasks and it was not impressive. i didn't expect it to be that good at agentic tasks

by skysniper

4/1/2026 at 4:44:52 PM

According to openrouter.ai it looks like StepFun 3.5 Flash is the most popular model at 3.5T tokens, vs GLM 5 Turbo at 2.5T tokens. Claude Sonnet is in 5th place with 1.05T tokens. Which isn't super suprising as StepFun is ~about 5% the price of Sonnet.

https://openrouter.ai/apps?url=https%3A%2F%2Fopenclaw.ai%2F

by hadlock

4/1/2026 at 5:21:28 PM

> the most popular model

It was free for a long time. That usually skews the statistics. It was the same with grok-code-fast1.

by NitpickLawyer

4/1/2026 at 5:54:10 PM

Exactly. When I read the headline I thought: "Ofc it is, its free."

by MaxikCZ

4/1/2026 at 6:03:03 PM

I should have clarified I didn't use the free version...

by skysniper

4/2/2026 at 3:53:11 AM

I used to use these various models for my claw-like and what they had a habit of doing is taking way more agent rounds and way more tokens to produce something that Sonnet would produce from far less. My total cost ended up being the same to do useful things.

by arjie

4/1/2026 at 4:53:24 PM

the real surprising part to me is that, despite being the cheapest model on board, stepfun is often able to score high at pure performance. Other models at the same price range (e.g. kimi) fails to do that.

by skysniper

4/1/2026 at 9:02:01 PM

Glm also has their subscription witch I would assume heavy users to use.

by gunalx

4/1/2026 at 5:43:18 PM

why do half the comments here read like ai trying to boost some sort of scam?

by dmazin

4/1/2026 at 10:17:22 PM

Because there's absolutely nothing stopping that from happening. There are bots on Reddit, there are of course bots on here, a VPN friendly site where you don't even need an email. But a lot of people don't want to admit it.

by Capricorn2481

4/1/2026 at 6:45:36 PM

Yet when I tried it it did absymal compared to Gemini 2.5 Flash

by grimm8080

4/1/2026 at 6:50:03 PM

what kind of tasks did you try?

by skysniper

4/1/2026 at 7:16:37 PM

Missing from the comparison is MiMo V2 Flash (not Pro), which I think could put up a good fight against Step 3.5 Flash.

Pricing is essentially the same: MiMo V2 Flash: $0.09/M input, $0.29/M output Step 3.5 Flash: $0.10/M input, $0.30/M output

MiMo has 41 vs 38 for Step on the Artificial Analysis Intelligence Index, but it's 49 vs 52 for Step on their Agentic Index.

by mgw

4/1/2026 at 7:34:31 PM

I will try and add it. But I doubt it works well because Mimo V2 Pro is beaten by stepfun even at performance leaderboard (price is not a factor in this leaderboard), so I expect MiMo V2 Flash to perform even worse.

by skysniper

4/2/2026 at 7:12:25 AM

Interesting, I found the pro version to be very capable.

If stepfun is even better, then Chinese models are getting really good.

by throwa356262

4/2/2026 at 5:05:20 AM

I'm not seeing Deepseek mentioned very often, which I've been using for Openclaw, very cheaply I might add, with great success. I think I loaded $10 to my account 2 months ago and I still havent needed to top up.

by clausewitz

4/2/2026 at 5:47:25 AM

Which deepseek exactly and what do you use it for? Just curious.

by wg0

4/1/2026 at 4:49:07 PM

It looks like Unsloth had trouble generating their dynamic quantized versions of this model, deleted the broken files, then never published an update.

by smallerize

4/1/2026 at 9:48:19 PM

This model is free to use, and has been for quite some time on OpenRouter. $0 is pretty hard to beat in terms of cost effectiveness.

by azmenak

4/1/2026 at 9:59:24 PM

yeah but i'm not using the free version for benchmark...

by skysniper

4/1/2026 at 5:32:14 PM

another thing from the bench I didn't expect: gemini 3.1 pro is very unreliable at using skills. sometimes it just reads the skill and decide to do nothing, while opus/sonnet 4.6 and gpt 5.4 never have this issue.

by skysniper

4/2/2026 at 7:18:42 AM

Gemini 2.5 pro was the best Gemini, it has gone downhill since

by throwa356262

4/2/2026 at 1:50:29 AM

this tracks with what i've seen too. gemini tends to 'overthink' tool calls - it'll reason about whether to use a tool instead of just using it. in my experience the models that are best at agentic tasks are the ones that commit to a tool call quickly and recover from failures, not the ones that deliberate forever and sometimes bail. would be interesting to see if the benchmark captures retry behavior since thats where cost-effectiveness really diverges

by zhangchen

4/1/2026 at 6:54:46 PM

Tried the free version on OpenRouter with pi.dev and it's competent at tool calling and creative writing is "good enough" for me (more "natural Claude-level" and not robotic GPT-slop level) but it makes some grave mistakes (had some Hanzi in the output once and typos in words) so it may be good with "simple" agentic workflows but it's definitely not made for programming nor made for long writing.

by sunaookami

4/1/2026 at 9:09:44 PM

What kind of creative writing are you doing? Fiction or non-fiction like blog posts?

by admiralrohan

4/2/2026 at 8:38:18 AM

Fiction. One of my "benchmarks" is giving the model a bunch of (self-made) text and having it simulate a 4chan thread about it. This tests tool use (calling the APIs), some skills, censorship and general creativity. Some models refuse every new turn after reading real 4chan threads ;) Claude is especially good at this surprisingly while GPT fails spectacularly and Gemini is just lazy (and barely usable since it's constantly overloaded). Qwen (coder-model from Qwen CLI, so Qween 3.5) is also very good but sadly not usable in Pi (they detect and block calls outside their CLI).

by sunaookami

4/1/2026 at 7:13:02 PM

it's actually pretty good at openclaw type of tasks for non technical users: lots of tool calls, some simple programing

by skysniper

4/1/2026 at 8:24:36 PM

Yeah this kind of stuff. I have no experience with OpenClaw though.

by sunaookami

4/1/2026 at 7:32:37 PM

i like StepFun 3.5 Flash, a good tradeoff

by grigio

4/1/2026 at 8:38:27 PM

people aren't just using Claude models any more? that's nice to see

by yieldcrv

4/1/2026 at 8:48:50 PM

well, I still want to use it but the first day i tried openclaw + opus, it costs me ~$500...

by skysniper

4/1/2026 at 10:02:31 PM

[dead]

by aplomb1026

4/2/2026 at 8:49:16 AM

[dead]

by hyperlambda

4/1/2026 at 9:35:11 PM

[dead]

by jghiglia

4/1/2026 at 9:41:51 PM

[dead]

by Caum

4/2/2026 at 4:33:21 AM

[dead]

by mtrifonov

4/1/2026 at 4:17:35 PM

I ran 300+ benchmarks across 15 models in OpenClaw and published two separate leaderboards: performance and cost-effectiveness.

The two boards look nothing alike. Top 3 performance: Claude Opus 4.6, GPT-5.4, Claude Sonnet 4.6. Top 3 cost-effectiveness: StepFun 3.5 Flash, Grok 4.1 Fast, MiniMax M2.7.

The most dramatic split: Claude Opus 4.6 is #1 on performance but #14 on cost-effectiveness. StepFun 3.5 Flash is #1 cost-effectiveness, #5 performance.

Other surprises: GLM-5 Turbo, Xiaomi MiMo v2 Pro, and MiniMax M2.7 all outrank Gemini 3.1 Pro on performance.

Rankings use relative ordering only (not raw scores) fed into a grouped Plackett-Luce model with bootstrap CIs. Same principle as Chatbot Arena — absolute scores are noisy, but "A beat B" is reliable. Full methodology: https://app.uniclaw.ai/arena/leaderboard/methodology?via=hn

I built this as part of OpenClaw Arena — submit any task, pick 2-5 models, a judge agent evaluates in a fresh VM. Public benchmarks are free.

by skysniper

4/1/2026 at 7:02:22 PM

Cheapest just isn't a very useful metric. Can I suggest a Pareto-curve type representation? Cost / request vs ELO would be useful and you have all the data.

by vessenes

4/1/2026 at 7:29:41 PM

TBH that was my initial thought too, but I found some problem using this approach:

Essentially I'm using the relative rank in each battle to fit a latent strength for each model, and then use a nonlinear function to map the latent strength to Elo just for human readability. The map function is actually arbitrary as long as it's a monotonically increasing function so it preserves the rank. The only reliable result (that is invariant to the choice of the function) is the relative rank of models.

That being said, if I use score/cost as metrics, the rank completely depends on the function I choose, like I can choose a more super-linear function to make high performance model rank higher in score/cost board, or use a more sub-linear function to make low performance model rank higher.

That's why I eventually tried another (the current) approach: let judge give relative rank of models just by looking at cost-effectiveness (consider both performance and cost), and compute the cost-effectiveness leaderboard directly, so the score mapping function does not affect the leaderboard at all.

by skysniper

4/1/2026 at 6:01:56 PM

Could you add a column for time or number of tokens? Some models take forever because of their excessive reasoning chains.

by johndough

4/1/2026 at 6:14:45 PM

both are shown in battle detail page already. Time is shown in Scores table. Number of tokens are shown in Cost details at the bottom of the Scores. (I thought most people just want to see cost in USD so I put token details at the bottom)

by skysniper

4/1/2026 at 8:19:53 PM

I would have liked aggregated results instead. Expanding 300 tables is a bit tiresome. But I guess that is easy with AI now. Here is a scatter plot of quality vs duration

https://i.imgur.com/wFVSpS5.png

and quality vs cost

https://i.imgur.com/fqM4edw.png

But I just noticed that my plot is meaningless because it conflates model quality with provider uptime.

Claude Haiku has a higher average quality than Claude Opus, which does not make sense. The explanation is that network errors were credited with a quality score of 0, and there were _a lot_ of network errors.

by johndough

4/1/2026 at 8:37:18 PM

> The explanation is that network errors were credited with a quality score of 0, and there were _a lot_ of network errors.

all network error, provider error, openclaw error are excluded from ranking calculation actually, so that is not the reason.

Real reason:

The absolute score is not consistent across tasks and cannot be directly added/averaged, for both human and LLM. But the relative rank is stable (model A is better than B). That is exactly why Chatbot Arena only uses the relative rank of models in each battle in the first place, and why we follow that approach.

a concrete example of why score across tasks cannot be added/averaged directly: people tend to try haiku with easier task and compare with T2 models, and try opus with harder task and compare with better models.

another example: judge (human or llm) tend to change score based on opponents, like Sonnet might get 10/10 if all other opponents are Haiku level, but might get 8/10 if opponent has Opus/gpt-5.4.

So if you want to make the plot, you should plot the elo score (in leaderboard) vs average cost per task. But note: the average cost has similar issue, people use smaller model to run simpler task naturally, so smaller model's lower cost comes from two factor: lower unit cost, and simpler task.

methodology page contains more details if you are interested.

by skysniper

4/1/2026 at 8:52:45 PM

I agree. If humans are allowed to pick the models, there will be an inherent bias. This would be much easier if the models were randomized.

by johndough

4/1/2026 at 10:35:41 PM

i added native plot and stats for aggregated results, on arena page. please check it out!

by skysniper

4/2/2026 at 9:19:41 AM

Nice! It would be even better if the model name was shown by default instead of having to hover, but I got the information that I wanted. In case you should be concerned about the aesthetics with too many model names, I can recommend the adjustText library in Python, which makes it so that labels do not overlap. Something similar probably exists in JS (or an LLM can just translate the relevant bits).

by johndough

4/1/2026 at 11:26:39 PM

The second chart depicts StepFun > Sonnet > Opus in quality?

by esafak

4/2/2026 at 12:18:56 AM

check out my reply, his chart is plotting the wrong metric (average quality score)

by skysniper

4/1/2026 at 7:29:29 PM

some kind of top-level metric like avg tokens/task would be useful. e.g. yes stepfun is 5% the price of sonnet, but does it use 1x, 10x or 1000x more tokens to accomplish similar tasks/median per task. for example I am willing to eat a 20% quality dive from sonnet if the token use is < 10% more than sonnet. if token use is 1000x then that's something I want to know.

by hadlock

4/1/2026 at 4:31:05 PM

Please don’t use AI to write comments, it cuts against HN guidelines.

by refulgentis

4/1/2026 at 4:45:36 PM

sorry didn't know that. Here is my hand writing tldr:

gemini is very unreliable at using skills, often just read skills and decide to do nothing.

stepfun leads cost-effectiveness leaderboard.

ranking really depends on tasks, better try your own task.

by skysniper

4/1/2026 at 4:56:49 PM

It’s too late once it’s happened. I was curious, then when I saw the site looked vibecoded and you’re commenting with AI, I decided to stop trying to reason through the discrepancies between what was claimed and what’s on the site (ex. 300 battles vs. only a handful in site data).

by refulgentis

4/1/2026 at 5:13:06 PM

Too late for what? For you? maybe. There are many others that are okay with it and it doesn't disminish the quality of the work. Props to the author.

by rat9988

4/1/2026 at 5:36:47 PM

> Too late for what? For you? maybe.

Maybe? :)

> There are many others that are okay with it

Correct.

> and it doesn't disminish the quality of the work.

It does affect incoming people hearing about the work.

I applaud your instinct to defend someone who put in effort. It's one of the most important things we can do.

Another important thing we can do for them is be honest about our own reactions. It's not sunshine and rainbows on its face, but, it is generous. Mostly because A) it takes time B) other people might see red and harangue you for it.

by refulgentis

4/1/2026 at 5:01:48 PM

all 300+ battle data are available at https://app.uniclaw.ai/arena/battles, every single battle is shown with raw conversional history, produced files, judge's verdict and final scores

by skysniper

4/1/2026 at 5:38:44 PM

Thanks! Is the judge an LLM? There's lot of references to "just like LMArena", but LMArena is human evaluated?

by refulgentis

4/1/2026 at 5:54:50 PM

> Is the judge an LLM?

Yes, judge is one of opus 4.6, gpt 5.4, gemini 3.1 pro (submitter can choose). Self judge (judge model is also one of the participants) is excluded when computing ranking.

> There's lot of references to "just like LMArena", but LMArena is human evaluated?

Yeah LMArena is human evaluated, but here i found it not practical to gather enough human evaluation data because the effort it take to compare the result is much higher:

- for code, judge needs to read through it to check code quality, and actually run it to see the output

- when producing a webpage or a document, judge needs to check the content and layout visually

- when anything goes wrong, judge needs to read the execution log to see whether partial credit shall be granted

if you look at the cost details of each battle (available at the bottom of battle detail page), judge typically cost more than any participant model.

if we evaluate with human, i would say each evaluation can easily take ~5-10 min

by skysniper

4/1/2026 at 5:58:40 PM

Fair enough, yeah, agent evals are hard especially across N models :/

Thanks for replying btw, didn't mean any disrespect, good on you for not getting aggro about feedback

by refulgentis

4/1/2026 at 6:17:51 PM

I appreciate honest feedback, best way to learn :)

by skysniper

4/1/2026 at 6:14:42 PM

>Other surprises: GLM-5 Turbo, Xiaomi MiMo v2 Pro, and MiniMax M2.7 all outrank Gemini 3.1 Pro on performance

This has also been my subjective experience But has also been objective in terms of cost.

by citizenpaul