Tweet épinglé
A dollar for $40 of DeepSeek V4 Pro usage?
Hard to say no to that.
English
Command Code
3K posts
Command Code
@CommandCodeAI
Command Code with taste; the first coding agent that observes how you write code and adapts to your preferences over time with meta neuro-symbolic AI `taste-1`.
San Francisco, CA Inscrit le Kasım 2023
2 Abonnements7.1K Abonnés
Command Code retweeté
@MiniMax_AI @Scobleizer Just shipped MiniMax M3 @CommandCodeAI with $1 Go plan 20x free usage and 50% off for next week.
Try the best coding harness for open models.
Command Code@CommandCodeAI
MiniMax M3 is live now on CommandCode! A frontier-class open-weight model with 1M context, frontier coding, agentic performance, and native multimodality Give it a try with our $1 Go plan with 10x free usage credits!
English
@88Vicarious @MrAhmadAwais @MiniMax_AI @Scobleizer Ask openclaw or hermes to use command code as a cli and they easily can. You don’t need a plugin or an mcp.
English
English
Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities
- Coding & Agentic Frontier: 59.0% SWE-Bench Pro, 66.0% Terminal Bench 2.1, 34.8% SWE-fficiency, 28.8% KernelBench Hard, 74.2% MCP Atlas
- MiniMax Sparse Attention scales context to 1M
- Natively Multimodal from Step Zero
API: platform.minimax.io
Token Plan: platform.minimax.io/subscribe/toke…
🚀New! MiniMax Code: code.minimax.io
Weights & Tech Report in ~10 Days
English
MiniMax M3 is now 50% off for the next week!
If you haven’t tried it yet, our Go plan starts at just $1 and includes $10 in usage credits.
English
50% OFF now — the drop just got sweeter!
x.com/CommandCodeAI/…
Command Code@CommandCodeAI
MiniMax M3 is now 50% off for the next week! If you haven’t tried it yet, our Go plan starts at just $1 and includes $10 in usage credits.
English
MiniMax M3 is live now on CommandCode!
A frontier-class open-weight model with 1M context, frontier coding, agentic performance, and native multimodality
Give it a try with our $1 Go plan with 10x free usage credits!
English
Yes $1 Go plan is cool. But more is coming.
Noon 12pm Monday, 1st June.
Command Code deal drop.
This one is gonna be crazy good!!
English
Weekend Supervised Vibe Coding
Achu - means `print` in Tamil
Built using @antigravity Google Flash 3.5 by burning my 1000 credits - then I pivoted to @CommandCodeAI DeepSeek Pro, after burning that, switched to raw @deepseek_ai pro in the terminal.
I am still testing :)
English
@tonyblu331 It’s included. Check the pricing page. And check /design in Command Code.
English
@CommandCodeAI Plans for MIMO Pro? It is really good at making UIs and SVGs.
English
This is getting ridiculous!
MiMo V2.5 is up to 99% off on Command Code.
Let's get it!
English
@_PaulVillarreal No. You will get the same deals and discounts. But top up has no free credits. For those you need to upgrade. We have Pro and Max plans.
English
@CommandCodeAI If I run out of credits and top up with $1 USD, will the balance work the same as before? Meaning, will it still have a value of $40 USD?
English
Command Code retweeté
It turns out that if you fix tool calling and add a taste layer on top of open models, you can get really good performance for very low cost.
@CommandCodeAI is how you do it.
Save money. Code more. I predict we'll all be using open models before long.
Ahmad Awais@MrAhmadAwais
BIG day for us!! @CommandCodeAI has crossed $1M in annual run rate, 1 trillion tokens of usage, with over 9K customers, just 24 days after our public beta launch. we believe this makes it the fastest-growing coding agent harness for open models. 3rd largest by usage. Command Code is built around two ideas: 1. open models should be production-grade for coding. 2. your coding agent should learn your taste. we're building for taste and developer experience. so instead of making a soup of thousands of models, we build for the best ones, open or closed. the goal: a coding agent that feels like an iphone, opinionated and with taste, not a random android or a windows phone with no taste. on the first idea: open models. we fixed the "open models aren't good enough at tool calling" problem. our research came down to two things, quality and speed, and both trace back to one root cause: broken tool-calls that open models produce, especially when you use a bad harness. open-model tool-call failures are not deep, they are a small finite set of contract mismatches. so we repair them, with zero token loss. what started as 4 repairs is now the largest repair layer in the space: 36k tool-call fix variants. i wrote the idea up openly¹ a few weeks ago, and it has quietly become a de facto way people fix open models. developers have either adopted Command Code or used the same idea to build repair harnesses for nearly every top coding agent. i take that as more meaningful validation than anything we could say about ourselves. on the second idea: taste. Command Code builds your coding taste into skills, learned from your accepts, rejects, edits, prompts, and the corrections you repeat. over time it drifts away from generic code and toward how you actually ship code. it learns continuously, and while it is early, the direction feels right. net effect: developers using Command are writing production-quality code on open models, 10x to 100x cheaper, without fighting tool calls, while building repo and team-wide coding taste that compounds. i believe these numbers are a consequence of getting those two things right. what's next. we've applied the same repair idea to ai design slop, and bundled a /design capability² so every developer can level up their design work. the early response has been great. we have a big roadmap ahead of us. the feedback we hear most is that Command Code feels fundamentally different: an approach built on taste and repair. we're going open source next month. today we're a cli at the core, and we're also launching a full-fledged gui app, sandboxed background agents, and cooking up something fun i can't wait to share. we're growing too, hiring in sf and remote worldwide. check open roles on my profile bio. try it now. npm i -g command-code if you like engineering deep dives on how we're doing all this, i've linked some relevant posts below.
English
adamlar 1$ yapmış kaçırmayın derim
Command Code@CommandCodeAI
This is getting ridiculous! MiMo V2.5 is up to 99% off on Command Code. Let's get it!
Türkçe
We're on the way of building something truly great 🔥
Ahmad Awais@MrAhmadAwais
BIG day for us!! @CommandCodeAI has crossed $1M in annual run rate, 1 trillion tokens of usage, with over 9K customers, just 24 days after our public beta launch. we believe this makes it the fastest-growing coding agent harness for open models. 3rd largest by usage. Command Code is built around two ideas: 1. open models should be production-grade for coding. 2. your coding agent should learn your taste. we're building for taste and developer experience. so instead of making a soup of thousands of models, we build for the best ones, open or closed. the goal: a coding agent that feels like an iphone, opinionated and with taste, not a random android or a windows phone with no taste. on the first idea: open models. we fixed the "open models aren't good enough at tool calling" problem. our research came down to two things, quality and speed, and both trace back to one root cause: broken tool-calls that open models produce, especially when you use a bad harness. open-model tool-call failures are not deep, they are a small finite set of contract mismatches. so we repair them, with zero token loss. what started as 4 repairs is now the largest repair layer in the space: 36k tool-call fix variants. i wrote the idea up openly¹ a few weeks ago, and it has quietly become a de facto way people fix open models. developers have either adopted Command Code or used the same idea to build repair harnesses for nearly every top coding agent. i take that as more meaningful validation than anything we could say about ourselves. on the second idea: taste. Command Code builds your coding taste into skills, learned from your accepts, rejects, edits, prompts, and the corrections you repeat. over time it drifts away from generic code and toward how you actually ship code. it learns continuously, and while it is early, the direction feels right. net effect: developers using Command are writing production-quality code on open models, 10x to 100x cheaper, without fighting tool calls, while building repo and team-wide coding taste that compounds. i believe these numbers are a consequence of getting those two things right. what's next. we've applied the same repair idea to ai design slop, and bundled a /design capability² so every developer can level up their design work. the early response has been great. we have a big roadmap ahead of us. the feedback we hear most is that Command Code feels fundamentally different: an approach built on taste and repair. we're going open source next month. today we're a cli at the core, and we're also launching a full-fledged gui app, sandboxed background agents, and cooking up something fun i can't wait to share. we're growing too, hiring in sf and remote worldwide. check open roles on my profile bio. try it now. npm i -g command-code if you like engineering deep dives on how we're doing all this, i've linked some relevant posts below.
English
@nimsbh_ai @MrAhmadAwais deepseek is facing a capacity issue atm.
i encourage you to try other models?
400 tps of step is awesome.
x.com/CommandCodeAI/…
Command Code@CommandCodeAI
Step 3.7 Flash is one of the fastest low-cost open models, now live in Command Code. • 400 tokens/sec 🍃 • 256K context window • 3 reasoning levels Sharp instruction-following and strong /design taste. $1 Go plan with 10x free usage credits, best way to try it.
English
@MrAhmadAwais @CommandCodeAI I am seeing a huge amount of latency increase every time a tool call is involved with these top tier oss models, and I currently find worse or equivalent to models like GPT 5.4 nano for my use cases. Am I missing anything here or is this how these models work.
English
how did we make deepseek outperform opus 4.7?
i've been thinking about why "open model bad at tool calling" is almost always a harness problem, not a model problem.
context: spent the two days looking at billions of tokens in @CommandCodeAI (tb open source ai cli) using deepseek. I ended up writing a tool-input repair layer. the trigger was watching deepseek-flash fail on the simplest /review run, every shellCommand and readFile call bouncing back with a raw zod issues blob, the model unable to recover because the error wasn't in a form it could read. by the end deepseek v4 pro was beating opus 4.7 6/10 times on our internal evals.
a few things i learned that feel general:
1/ the failure modes aren't random they're a small finite compositional set.
across deepseek-flash, deepseek v4 pro, glm, qwen, the same four mistakes repeat almost exactly:
- sending `null` for an optional field instead of omitting it
- emitting `["a","b"]` as a json *string* instead of an actual array
- wrapping a single arg in `{}` where the schema expected an array (an "empty placeholder")
- passing a bare string where an array was expected (`"foo"` instead of `["foo"]`)
four repairs, ~30-100 lines each, ordered carefully (json-array-parse must run before bare-string-wrap or `'["a","b"]'` becomes `['["a","b"]']`). that is the whole catalogue. when i hear "this open source model can't do tool calls" i now assume one of those four, and so far that's been right ~90% of the time.
2/ the funniest failure mode is also the most revealing.
deepseek-flash, when asked to edit or write a file, sometimes emits the path as a *markdown auto-link*:
filePath: "/Users/x/proj/[notes.md](http://notes. md)"
our writeFile tool obediently trued creating files literally named `[notes.md](http://notes .md)` until we caught it. this is not a hallucination. it's the post-training chat distribution leaking through the tool boundary the model has been rewarded for auto-linking in conversational output, and is applying that prior in a context where it makes no sense. the fix is two regex lines that unwrap only the degenerate case where link text equals url-without-protocol real markdown like `[click](https://x .com)` passes through untouched.
this is also conditioning of their own tools during RL which were different from all other tools we write and ofc can't predict.
"tool confusion" is a more useful frame than "capability gap." the model knows how to format a path. it just hasn't been told clearly enough that this path is going to fopen, not into a chat bubble. so we encode that hint at the schema level `pathString()` instead of `z.string()` and the leak is plugged for every path field at once.
3/ the design choice that mattered was inverting preprocess-then-validate to validate-then-repair.
my first attempt was the obvious one: a preprocessing pass that normalized inputs (strip nulls, parse stringified arrays, etc.) before zod ever saw them. it broke immediately, writeFile content that *happened* to be json-shaped got rewritten before it hit disk. silent corruption, easy to miss in a smoke test.
then i made it less greedy
- parse the input as-is. if it succeeds, ship it. valid inputs are never touched.
- on failure, walk the validator's own issue list. for each issue path, try the four repairs in order until one applies.
- parse again. on success, log `tool_input_repaired:${toolName}`. on failure, log `tool_input_invalid:${toolName}` and return a model-readable retry message.
the structural insight here is: when you preprocess, you encode a prior about what's broken. when you let the validator complain first, the schema is the prior, and you only spend repair budget at the exact paths the schema actually disagreed at. the validator is doing the work of localizing the bug for you. it's the same shape as cheap-then-careful everywhere else try the fast path, fall back on evidence.
(this also gives you per-tool telemetry for free. you can watch repair rates per (model, tool) and notice when a model regresses on a specific contract before users do.)
4/ shape invariants and relational invariants need different fixes.
the four repairs above all handle shape problems wrong type, missing key, wrong container. but read_file had a *relational* invariant: "if you provide offset, you must also provide limit, and vice versa." deepseek kept calling `readFile({ absolutePath, limit: 30 })` and getting an `ERROR:` back. you can't fix this with input repair, because each field is independently valid the bug is in the relationship between them.
so i taught the function the model's intent instead. `limit` alone → `offset = 0`. `offset` alone → `limit = 2000` (matches common read tool ops default). then surfaced the decision back to the model in the result:
"Note: limit was not provided; defaulted to 2000 lines. To read more or fewer lines, retry with both offset and limit."
no `Error:` prefix, so the tui doesn't paint it red. the model sees what we picked and can self-correct on the next turn if our guess was wrong. transparency over silent magic wins big.
repair where you can. extend semantics where you can't. surface the choice either way.
zoom out:
a lot of what looks like model capability is actually contract design. a strict schema is a choice with a cost it filters out noise, but it also filters out recoverable noise from any model that hasn't memorized the exact json contract you happened to pick. the largest commercial models eat that cost invisibly and are linient on tool calling because they've seen enough of every contract during pretraining; open models pay it loudly and get dismissed for it.
the harness is where you mediate between distributions. four small repairs (i'm sure more to follow as we have three more merging today), two regex lines for auto-links, one relational default, one prefix change. the model didn't change. the contract got more forgiving in exactly the places it needed to be.
deepseek v4 pro now beats opus 4.7 6/10 times on our internal evals.
imo "skill issue" applies to the harness more often than the model.
Ahmad Awais@MrAhmadAwais
Wow I just made DeepSeek V4 Pro beat Opus 4.7 6/10 times in our internal evals by auto repairing many of its quirks in tool calling. It’s performing super solid for such a cheap model.
English
@AlexGonchX @MrAhmadAwais avg developer using Command Code is doing 106M tokens. looking at the last 28 days of data since month isn't over yet.
English
@MrAhmadAwais @CommandCodeAI that's like 5m tokens a day per user. who's actually running agents that hard
English
Big announcement today.
We raised $5M to build the first coding agent that continuously learns your coding taste.
Introducing Command Code.
$ npm i -g command-code
`cmd` learns how you write code.
Every accept, reject, edit is a signal.
Code 10x faster. Review 2x quicker.
English