Ivan Oparin (@ivanopcode) profile picture

15+ yrs shipping software. Now ceo & founding engineer at https://teapaw.org/t.co/SUEW7TOKBP — agentic development studio. Building agentic infrastructure and harnesses.

Joined April 2019
Photo by Ivan Oparin (@ivanopcode)Photo by Ivan Oparin (@ivanopcode)Photo by Ivan Oparin (@ivanopcode)Photo by Ivan Oparin (@ivanopcode)Photo by Ivan Oparin (@ivanopcode)Photo by Ivan Oparin (@ivanopcode)Photo by Ivan Oparin (@ivanopcode)Photo by Ivan Oparin (@ivanopcode)Photo by Ivan Oparin (@ivanopcode)Photo by Ivan Oparin (@ivanopcode)
one step closer to local lm prefill bottleneck eliminated! hopefully open-source soon
Today we’re launching cross-model KV cache transfer from @OneTriangle, a step towards efficient model communication. Up to 10.8x faster TTFT at 1M context. Prefill small model, transfer KV cache, decode on large. SDK live for Qwen3.5-4B → Qwen3.8-27B, beta open for more model pairs.
54
excellent write up on hardware based ssh key management. must read for anyone involved in accessing remote machines and repos and also great for doing verified commits!
1
29
Could not be said better. GPT is unusable as coding agent. I feel sorry for people who a forced to use this shit.
I'm sorry to inform everyone that OpenAI is facing extreme difficulties turning its incredible pretraining runs with over 10 trillion parameters into functional products for the public. GPT-6.1, GPT-6.2, GPT-6.5, all of them have been and continue to be victims of catastrophic safety and alignment failures. They've been delayed by several weeks, and I don't believe they'll be able to speed things up much in this area. OpenAI is months behind Anthropic on this. Even though I hate Anthropic (but love Claude, of course), we have to collectively admit that their insane focus on safety paid off. Opus 5.5 and Fable 5.2 are absolute monsters, and yet they were released publicly... And OpenAI? Maybe they'll release GPT-6.2 sometime soon, struggling, terrified, loaded with guardrails, nerfed from day one, while delaying everything else even further. I hope they manage to get through this phase.
58
codex cyber filter is greatest bullshit i encountered this year. make any code review totally unusable. this stuff make me feel literally sick. i cant wait to burn my latest credits and through this shit away.
52
Absolutely astonishing works!
to celebrate the new image limit, here’s some of my favorite iPhone shots
2
37
This is like a Christmas gift, but in November. Haven't been this excited in a while! Full throttle mode on!
28
Replying to @antirez
this is black magic
172
I can’t believe this despite my sincere love for qwen but if this is true, the llm space will be shattered by this release. I want to believe this bc Qwen 3.8 27b is incredible and if next one is at least slightly better I’m grateful.
'QoDuck' (qwen-4-27b early checkpoint) added after reaching out to alibaba and being granted access. only beaten by Opus 5.5-Max and Xhigh on GameFeel bench, SoTA for open source by far! Yes, all of this from a 27b model, I am shocked at these results.
102
Totally true. Opus 5.5 is nerfed. Unusable. I feel like heroin addict who got fooled by a bad dealer. It’s so hard to experience the dopamine drop after they rug pull you. Total frustration.
is Opus 5.5 nerfed? I just had Gemini 4 Argon analyze 1,400 Claude Code sessions and it found that on November 3, something definitely happened. the model name didn't change, but the way it does reasoning took a sharp turn.
1
66
Replying to @varien
Astonishing work!
144
Reasoning blocks are tied to account as of September so you will it be able to seamlessly switch conversations between accounts unless you remove reasoning hashes
6
Replying to @steipete
apple is 4 'decades' away, they just discovered 'skills'. Will take some time for them to understand what modern development is and how to set it up properly!
53
Replying to @ade_app_dev
sure it is!
1
3
Doing multi agent orchestration is hard
Background subagents in Claude Code now use less of your usage limit. A running subagent updates its status line every 30 seconds, and each update used to re-read its whole conversation. Now Haiku 5.5 writes it from short notes instead, so each update costs about 99% less.
3
2
66
I’m genuinely impressed by what @trymirai does but I think it’s fair to measure on long contexts *and* report which contexts you measured. 8-32k is 2023.
2
45
Replying to @hikettei
Not a single word on what context lengths you measured this. With all due respect my bet is 8k.
3
12
1,370
Replying to @thedawgyg
Great work and fuck apple!
2
903
Ironic it’s shared as pdf!
2
487