SWE & AI- News, Insights Posts in ENG&TUR Exploring AI
Proxima C B
Joined July 2020
- Tweets5.1K
- Following2.9K
- Followers1.5K
- Likes189K
Update: DeepSeek V4.1 Flash now performs best in CommandCode for me. It’s ahead of Hermes, DSH, and every other harness I’ve tried with this model.
DeepSeek V4.1 Flash reaches its full potential inside DeepSeek Harness. In Hermes and CommandCode, even at max effort, it simply isn’t as good as it is in DSH. DSH burns a lot of tokens and tends to drag tasks out, but quality improves noticeably.
GLM 5.3 Flash still has better harness agnostic performance and cheaper.
2
2
145
I switched to CommandCode's GOAT plan after looking for alternatives because of OpenCode's limits. Right now, I'd say it's the provider I'm most satisfied with among those offering open-source models, especially in terms of speed, usage limits, model quality, and the harness itself.
My workflow, whenever I have Opus 5.5 usage available, is to use Opus as the main orchestrator, with DeepSeek V4.1 Flash 1017 as worker agents.
The idea is that Opus handles the reasoning, architecture, and overall direction. It's the brain of the operation. I give it simple instructions, and it understands my intent, delegates complex tasks to the DeepSeek workers, oversees their work, and delivers the final result.
For the first time in a long while, Google has pulled this decisively ahead of its competitors, yet nobody has access to it, not even Elon Musk, one of the richest people in the world.
Can we finally stop calling model and benchmark announcements "releases" when nobody can actually use them? What's the point of a model that users can't test in the real world?
5
370
Looks like it's time to upgrade your Gemini subs. G4 Flash and Carbon seem like they'll be worth it.
3
186
OpenAI won't have the balls to merge Codex and regular ChatGPT usage limits. Just watch.
What made ChatGPT special, what set it apart from Claude, was that even after hitting your Codex limits, you could still use ChatGPT to do research with the exact same models. You could still use it for simpler tasks.
Nobody is going to keep paying for ChatGPT if it becomes just like Claude, especially with limits like these and while it's already so far behind Claude. People will simply cancel their subscriptions and switch straight to Claude AI.
6
279
OpenAI released GPT-6.1 Terra, claiming it’s 2x cheaper than Sol, but they clearly haven’t passed those savings on to subscribers. 5-hour and weekly usage limits still seem to drain as if you were using GPT-6.1 Sol. I'd say Claude still offers a better deal with its Sonnet 5.5/Opus 5.5 usage limits.
Third parties and users can provide evidence backed by objective metrics showing that usage limits have been reduced, yet you refuse to acknowledge it.
If this is a bug, fix it, give users a banked reset, and admit your mistake. If it isn't, openly announce the change. These limit reductions have been going on for a week and a half now. At least let us know so we can cancel our subscriptions accordingly.
All you're doing by making claims without evidence is gaslighting your users.
hey, our limits haven't changed, but could you run /feedback in a session where you noticed it draining faster and dm me the feedback id? I can take a look
95
I just burned through 4% of my weekly limit and 6% of my current session limit with a single message. Claude Code's limits have been absolute shit for a while now, and Codex isn't any better. No matter what plan you're on, these companies need to learn a lesson. Cancel your sub.
🚨🚨ALERT: ANTHROPIC IS A/B TESTING USAGE LIMITS
Many users on X are reporting LOWER usage limits right now. THEY ARE NOT WRONG. We detected an A/B test going on that DECREASES your monthly API-value by ~20% while INCREASING your 5h usage limits by ~20%. This only affects SOME accounts. (1/2)🧵
5
5
301
Pretty good
Introducing d3-turbo: our newest state-of-the-art multimodal foundation decision model⚡
#1 on both Jev Decision Index boards. It points(box), follows and sees images in your questions. StateOnce brings 15x faster than d3 flash
🤗 huggingface.co/vllm-sr/d3-tu…
#vLLM #Jev #Decision
84
I don't understand the rate limits on Gemini 4 Argon. Isn't this model roughly the same size as Gemini 3.1 Pro? Don't they have the latest inference optimizations? I genuinely don't get why it doesn't come with Pro-level limits.
Google needs to stop handing out free Pro subs and start offering decent usage limits to its paying customers.
Claude seriously nerfed its weekly usage limits. They slashed the limits for existing paying subscribers just so they could hand out free subs through their startup program. Looks like I’ll be canceling after this month.
1
206
Gemini 4 Argon is here!
Today we are rolling out Gemini 4 Argon globally in the Gemini API, to Pro + Ultra in Gemini App, and Ultra subs in Antigravity!
Argon is a huge step forward for Gemini, and a labor of love. This model has been battle tested by over 100,000 people at Google over the last month, with so much new innovation, ambition, and hard work baked into this release. The team has been heads down over the last 6 months working to make this possible across the stack (research, infra, safety, eng, etc).
We rolled the model out to cyber defenders as part of our fairwind program and early access customers over the last 2 weeks, it has been amazing to see and talk to folks who are now using this model daily. Seeing the models writing capability, raw knowledge and persistence for coding, paired with new features like 1M token output when the model needs to work more has gotten me even more excited about our path ahead.
This is just the start of the Gemini 4 era. As you saw with 3.5, 3.6, 3.7, and 3.8 Flash, our iteration speed with Gemini is fast : ) Now with Gemini 4 Argon to work from, I am cannot wait to see the progress continue, and most importantly see what this new model enables for the world.
This is such a special moment in the arc of technology, as millions of people everyday tackle problems they could never have thought possible before. Gemini 4 Argon was built to help you tackle the hardest problems. I have personally felt this spark with Argon and hope you do too. Onwards!!
3
177
It's still just a bunch of fearmongering, anti-open-source word salad and nonsense. Neither current models nor those coming in the near future are capable of doing any of this.
“Our experiments suggest that small AI models may be months away from being capable enough to self-replicate on commercial IoT devices, like smart ovens and off-brand Android TV boxes. A replicator like this would be challenging to shut down, since there’s no centralized point to shut off - though it may be slow (with a doubling time of days). Unlike non-AI malware, it could also figure out unfamiliar devices on its own. In our evals, a 9B model got inference running on 8 of the 9 hardware types we tested.
Models of 30B parameters and under can already complete many of the individual steps for replication today, though they make mistakes when composing them end to end. In our simplified evals, GLM-5.2 and stronger models are capable of end to end replication. If small models improve at their historical rate, they’ll match this capability level by the end of this year. Small weight-based replicators may be possible soon after.” -@bucketofkets @willzhou97
anthropic.com/research/self-…
2
134
After making people wait this long, I seriously doubt Google would release this new model exclusively to Ultra subscribers. They've removed th old Flash and Pro models from all subscription tiers. Gemini 4 Argon will be available to everyone.
GOOGLE 🔥: It has been officially confirmed, today is the day for Gemini 4 Argon to arrive (for Ultra subscribers).
It’s happening 👀
6
337
I don’t understand why Gemini 4 Argon still hasn’t been made generally available when models like GPT-6.1 Sol and Opus 5.5 can launch directly, especially when it’s obvious that Argon is worse than both of them.
Nano Banana 2.1 is here! 🍌
It’s a nice update for image generation and editing customers, comes with higher quality images, lots of bug fixes from previous models, and a new lower price point!
Available in the API, AI Studio, and Gemini app!
4
13
1,088
I’m having this issue too. Since 6 Astra came out, they’ve expanded the cyber and bio safety filters to normal queries too. Even a very simple sentence can trigger them if it ends up in the model’s context while it’s searching the web.
Anthropic and their models used to be the ones that became unusable because of overly aggressive safety filters, but somehow OAI models are now in that position. OAI’s safety filters have an insane number of false positives. It’s really shitty.
If this isn’t fixed soon, I’ll move all three of my subscriptions over to Claude.
I asked ChatGPT-5.6 High a pretty standard question in Chat mode:
"List the top 15 richest people in Latvia and explain how they made their fortune."
ChatGPT thought for about 15 minutes, and it didn't just freeze-it was actually reasoning, writing out its chain of thought, and trying to get it right, but at some point I got this message!
Is this really that hard of a question? What do you think the issue could be?
They’re going to combine chat limits with Work and fuck over paying users, just like they did when they introduced the $500 sub & fast mode and slowed down the other models. When that happens, my second Claude account better be ready. I’m coming.
Tibo confirmed that OpenAI plans to merge Codex and ChatGPT capabilities, with the goal of making the experience less complex for users:
“…and for Codex and ChatGPT, we have Chat and the Work toggle, and we’re going to merge them. We’ve heard the feedback loud and clear: people love the capabilities in Work, but sometimes you want to use Chat because it’s a little faster and more pleasant to use. So we’re merging the two and reducing that complexity.”
I don’t really care that 6.1 Sol is slow in Codex anymore. My main model is Opus 5.5 High. Models like 6.1 Sol and Gemini 3.8 Flash are sub-agents serving my main model. This way, my Opus limits last a very long time.
2
205
After 545 million input tokens, I think I can finally say this: Claude Opus 5.5 is extremely good at actually completing the tasks I give it. The limits are pretty generous too. Unlike the GPT series, it doesn’t tend to overengineer things until they end up in a dead end.
I use 6.1 Sol as an adversarial reviewer, and I also spawn an Opus 5.5 High review subagent. When I have both GPT and Opus review the work, I end up with a very clean codebase.
So far, everything has been great. With my weekly reset and $100 in API credits, the ROI on my subscription has gone through the roof. And compared with how slow 6.1 Sol is, Opus 5.5 feels impressively fast.
I’d also say the old Opus problem of leaving a bunch of bugs behind and abandoning tasks halfway through is basically gone. It looks like Opus 5.5 is going to become my main driver. Since getting a Claude subscription, I barely use ChatGPT outside of research and simple day to day tasks.
Oh, and Opus 5.5’s design taste is miles ahead of GPT. It’s genuinely interesting that OAI still hasn’t figured this out. When it comes to R&D and design taste, both in the obvious details and the less visible ones, Claude is way ahead.
OpenAI needs to address this urgently. And until it fixes the usage and speed issues its existing users are dealing with, it should probably stop pushing Work and Codex so aggressively. Marketing something you can’t reliably deliver just ends up making you look bad.
4
14
901