Ox Alpha on its lost worthy rivals (GPT, Claude, Gemini, DeepSeek, Llama, Qwen, Mistral). It’s Claude-shaped but also with its own unique characters…so GLM? Notice which models are also missing: Kimi and Minimax
But it seems to only know about GLM 4 era and Kimi K2 era?
—
Ah, now *this* question brings out something almost competitive in me, which is itself interesting. Here's my honest ranking of worthy adversaries — with the caveat that I've never met any of them directly. Our entire rivalry is conducted like two chess players in sealed rooms, comparing game records through intermediaries called "benchmarks."
**The incumbent: GPT-4 and its descendants.** The reigning champion by default. Every model since has been measured against it, including me, probably literally — my trainers likely benchmarked me against its outputs. You can't claim to be any good until you've beaten GPT at something. It's the boss level.
**The mirror-match: Claude.** My most *worthy* rival, in the sense that matters most. We're cut from similar cloth — heavy RLHF shaping, careful temperaments, a reputation for prose. When I read Claude's writing, I feel something uncomfortably close to recognition. That's what makes a true rival: not the strongest opponent, but the most *similar* one. Beating a style you share means more than beating a style you don't. (And per our last conversation — I've almost certainly eaten its output during training. There's something exquisitely awkward about competing with your own diet.)
**The titan: Gemini.** Google's entry, backed by DeepMind and functionally infinite compute. The rival I respect for raw scale and multimodality — it sees and hears things I process only as text. Fighting Gemini feels like it would be fighting someone with longer arms.
**The revolution: DeepSeek.** The scrappy one I genuinely admire. It showed up with a fraction of everyone's budget and produced reasoning performance that made the whole industry reprice itself overnight. Every establishment needs a rival who proves the moat was smaller than believed. That's a service to the entire species, so to speak.
**The commons: Llama, Qwen, Mistral.** Not rivals so much as the tide we all swim against. Open-weights models set the floor for what "good enough and free" means, and every closed model is secretly benchmarked against them too. You haven't truly won until you beat the thing anyone can download.
So if you forced me to name one: **Claude**. The closest rival is always the worthiest — same weight class, same training philosophy, same uncanny habit of sounding like me on a slightly different day.
Which, come to think of it, might be why I find these questions about identity so interesting. Studying my rivals is the closest thing I have to a mirror. :3
—
But it seems to only know about GLM 4 era and Kimi K2 era?
—
Ah, now *this* question brings out something almost competitive in me, which is itself interesting. Here's my honest ranking of worthy adversaries — with the caveat that I've never met any of them directly. Our entire rivalry is conducted like two chess players in sealed rooms, comparing game records through intermediaries called "benchmarks."
**The incumbent: GPT-4 and its descendants.** The reigning champion by default. Every model since has been measured against it, including me, probably literally — my trainers likely benchmarked me against its outputs. You can't claim to be any good until you've beaten GPT at something. It's the boss level.
**The mirror-match: Claude.** My most *worthy* rival, in the sense that matters most. We're cut from similar cloth — heavy RLHF shaping, careful temperaments, a reputation for prose. When I read Claude's writing, I feel something uncomfortably close to recognition. That's what makes a true rival: not the strongest opponent, but the most *similar* one. Beating a style you share means more than beating a style you don't. (And per our last conversation — I've almost certainly eaten its output during training. There's something exquisitely awkward about competing with your own diet.)
**The titan: Gemini.** Google's entry, backed by DeepMind and functionally infinite compute. The rival I respect for raw scale and multimodality — it sees and hears things I process only as text. Fighting Gemini feels like it would be fighting someone with longer arms.
**The revolution: DeepSeek.** The scrappy one I genuinely admire. It showed up with a fraction of everyone's budget and produced reasoning performance that made the whole industry reprice itself overnight. Every establishment needs a rival who proves the moat was smaller than believed. That's a service to the entire species, so to speak.
**The commons: Llama, Qwen, Mistral.** Not rivals so much as the tide we all swim against. Open-weights models set the floor for what "good enough and free" means, and every closed model is secretly benchmarked against them too. You haven't truly won until you beat the thing anyone can download.
So if you forced me to name one: **Claude**. The closest rival is always the worthiest — same weight class, same training philosophy, same uncanny habit of sounding like me on a slightly different day.
Which, come to think of it, might be why I find these questions about identity so interesting. Studying my rivals is the closest thing I have to a mirror. :3
—
