Worth highlighting the QOI Image format (qoiformat.org) showing that you can also get significant image compression benefits with a simple 1-page specification [1]:
> QOI is fast. It losslessly compresses images to a similar size of PNG, while offering 20x-50x faster encoding and 3x-4x faster decoding.
> QOI is simple. The reference en-/decoder fits in about 300 lines of C. The file format specification is a single page PDF.
Because they want to optimize it for their models and don't want to be blocked by waiting for PRs to merge or be rejected.
There's plenty of reasons to start your own fork that you have full agency of, as long as the OSS License is maintained anyone will be able to benefit from any new features they want to make use of.
Everyone has to rely on Iran's satellite images to find out what's happening because Trump is censoring the war by forcing a blackout of commercial satellites from providing any images.
I said speed was great, Cerebas and Groq can provide better performance, likewise Fast versions of Cursor's Composer and Claude.
The reported speed like benchmarks is only a reported number on paper, we'll see how it holds up in real world usage, so far OpenRouter is only reporting 73tps
It absolutely matters, especially when done in unison like this.
Cancelling ChatGPT sends a signal that you don't agree with weaponizing AI. Switching to Claude says you support Anthropic's principled stance against it. If you have a strong opinion either way, today is the day to vote with your wallet.
Dismissing every small action as meaningless is just apathy and how nothing ever changes.
Perfect timing - Had already cancelled my Claude sub over their OAuth ban in external tools and was about to pick up a Codex sub as the next best alternative.
Ended up renewing my Claude sub today instead. Principled stances matter and I no longer trust OpenAI to be trustworthy custodians of my AI History.
At this point I'd go far to say I wouldn't trust any company with my AI history that caves to DoD demands for mass domestic surveillance or fully autonomous weapons.
Your AI will know more about you than any other company, not going to be trusting that to anyone who trades ethics for profits.
Had Cancelled my Claude sub after they banned OAuth in external tools, but just renewed it today after seeing their principled stance on AI ethics - they matter more when they hurt profits, happy to support them as a Customer whilst they keep them.
Servo isn't a JS engine. Do you mean why didn't they abandon their mission statement of developing a truly independent browser engine from scratch, abandon their C++ code base they spent the last 5 years building, accept a regression hit on WPT test coverage, so they can start hacking on a completely different complex foreign code-base they have no experience in, that another team is already developing?
I consider HuggingFace more "Open AI" than OpenAI - one of the few quiet heroes (along with Chinese OSS) helping bring on-premise AI to the masses.
I'm old enough to remember when traffic was expensive, so I've no idea how they've managed to offer free hosting for so many models. Hopefully it's backed by a sustainable business model, as the ecosystem would be meaningfully worse without them.
We still need good value hardware to run Kimi/GLM in-house, but at least we've got the weights and distribution sorted.
Sure he's still one of the top players, but he's not as strong this year and OP is suggesting he still has an edge against the GOAT, who this year:
- Has won Freestyle WC
- Has won SCC
- Has won 2x Titled Tuesday's
- Has won a Freestyle Friday
Hikaru can snipe a win off Magnus here and there, but I don't think there's any time control or format where he could win a long series of chess matches against Magnus.
Hikaru is either in a slump or his skill is starting to age: hasn't won Titled Tuesday since November, hasn't won Freestyle Friday this year, came last in Speed Chess Championship, etc.
We'll see how well he does in Candidates this year to see if he's still a top contender. Although I do believe this is his last chance to fight for the world title.
Using a coding plan, haven't noticed any throttling and very happy with the performance. They publish the quotas for each of their plans on their website [1]:
Really looked forward to this release as MiniMax M2.1 is currently my most used model thanks to it being fast, cheap and excellent at tool calling. Whilst I still use Antigravity + Claude for development, I reach for MiniMax first in my AI workflows, GLM for code tasks and Kimi K2.5 when deep English analysis is needed.
Not self-hosting yet, but I prefer using Chinese OSS models for AI workflows because of the potential to self-host in future if needed. Also using it to power my openclaw assistant since IMO it has the best balance of speed, quality and cost:
> It costs just $1 to run the model continuously for an hour at 100 tokens/sec. At 50 tokens/sec, the cost drops to $0.30.
> QOI is fast. It losslessly compresses images to a similar size of PNG, while offering 20x-50x faster encoding and 3x-4x faster decoding.
> QOI is simple. The reference en-/decoder fits in about 300 lines of C. The file format specification is a single page PDF.
[1] https://qoiformat.org/qoi-specification.pdf