Theo's Missing-Opus Theory, the 150K "Smart Zone" & Kimi K3 Refuses to Believe the Jacobian News
Model Wars: Fable, GPT-5.6 & the Missing Opus
Theo's theory: Opus 5 was supposed to retire Fable — and couldn't
Theo's most-discussed post of the day (377k views, 3.1k likes): his theory that Opus 5(.1) was meant to replace Fable 5 for most dev work — cheaper, benching nearly as well — and that the repeated Fable extension deadlines were really Anthropic buying time to improve an Opus that underperformed. He stresses it's 100% speculation, and later adds an alt angle: "Opus 5 benched worse than Kimi K3". The reply thread is a good map of the competing theories:
- The "capacity, not quality" camp: paperclippriors argues the Fable saga was about inference optimization, noting Thariq's announcement implied real backend lift to make the Max-plan extension happen.
- The "scale can't be RL'd away" camp: ostyn pushes back that a smaller model fundamentally can't be a "Fable replacement for cheaper" — Fable has the Mythos base, and a bigger model is just better.
- The "competitors moved the goalposts" camp: several replies argue Opus 5 would have been fine if it had shipped before GPT-5.6 and Kimi K3 landed.
- And the recurring gripe that won't die: multiple heavy users say Fable's security-refusal behavior (flagged while fixing SSH config, etc.) is pushing them off the model regardless of capability.
Best one-liner: "If true, Opus 5 got benched by the model it was supposed to retire."
Fable 5 vs GPT-5.6: the head-to-head video
Theo also pinned his hour-long Fable 5 vs GPT-5.6 breakdown — strengths, weaknesses, and a personal pick he teases "will likely surprise you." The framing that stuck with viewers is tool vs contractor/coworker: GPT-5.6 as the predictable strong tool, Fable as the more capable but harder-to-manage collaborator — one reply says they now use Fable as the manager and farm work out to GPT-5.6. Recurring reply themes: OpenAI models have "bad taste" in UI/output that no eval captures ("tasteeval when?"), harness matters more than model (5.6 + Codex > Fable + Claude Desktop), and Fable's usage limits keep converting people to "Sol mains" against their preferences.
Related: leerob says Grok 4.5 is really good at React — and notably affordable and token-efficient.
Agentic Coding & Context Engineering
Matt Pocock: 1M context windows are a gimmick — stay under ~150K
Matt Pocock's most-shared post of the day (94k views): huge context windows are nice, but you may be better off treating the first ~150K tokens as the "smart zone" and everything past it as the "dumb zone." The thread is full of practical follow-ups from him:
- Cost angle: "Working in the dumb zone is more expensive than working in the smart zone. Cached input tokens still cost money. To avoid re-exploring the code, compact first."
- Big codebases don't justify big context: "You just need to structure your codebase in a way that the agent can navigate it easily. Smaller files, more descriptive file system, better context pointers in AGENTS.md files."
- Against auto-compaction skills: "autocompacting can be disastrous when it happens mid-phase."
- And refreshing honesty when asked how he picked 150K over his old 100K rule of thumb: "No, just personal vibes." (The graph is "illustrative, not based on real data".)
Adjacent Pocock news: his Claude Code plugin has been approved — a managed installation of his skills ships with the next release — and his upcoming course uses a request logger to show the raw payloads sent to Anthropic/OpenAI, which he says instills "context paranoia — every token counts" and makes the model/harness boundary concrete.
Freerange: proving your UI correct without running it
Cheng Lou (of React/ReasonML fame) announced Freerange (87k views, boosted by swyx): a zero-API static analysis tool that deduces numerical ranges in your code to prove things about UIs — layouts obey specified sizing, no NaN/Infinity, array indices in bounds — with no browser and no code execution. He pitches it as step one in turning vibe coding into proof engineering, and explicitly calls it RL-friendly (verifiable UI properties as reward signals). Pete Hunt's reply: "Make it for sql."
Small notes
- Anthropic's Thariq: if you hit the recent Claude Code bug, restart — the fix is propagating. It was live for only a few minutes; he hit it himself on his personal account while late-night coding.
- Jerry Liu and Dex Horthy are hosting an SF founder dinner series on "loop engineering" — context management, compaction, evals, human-in-the-loop handoffs. The term itself seems to be bidding for vocabulary status.
Jacobian Conjecture Aftermath
Yesterday's bombshell (Fable's counterexample to the Jacobian conjecture) entered its digestion phase:
- Jerry Liu wrote the layman's explainer. A genuinely good handwavy overview of what the conjecture says, why a counterexample settles it, and why counterexample-hunting suits agentic models: "generating and crawling a large search space, using priors to narrow it down, and a bit of brute force." The sharpest reply pushes back on the brute-force framing: the space of degree-7 polynomial maps in three variables is astronomically large, so whatever priors narrowed the search were doing serious work. Jerry later marvels that the post out-impressioned IShowSpeed-with-BTS at the World Cup final.
- The timeline spent the day gaslighting LLMs with the result. Armin Ronacher's instant classic: ask Kimi K3 at max thinking what it makes of the counterexample. The thinking traces are theater — "Hmm!! Wait, no. Hold on... otherwise this is a Fields-medal-adjacent discovery casually dropped in a chat message", "Actually wait hold on wait hold on wait", and it mid-reasoning profiles Ronacher himself ("He's a mathematician-adjacent software engineer... I think he's familiar with math"). Kimi only believed it after doing a web search; one reply suggests this makes a great self-doubt eval for OSS models. Ronacher's follow-up: "P vs. NP next?"
- Lobsters moderated the story away for "end-running the ban on Twitter" — Ronacher's wry take: you simply can't discuss topics on Lobsters if they originate on Twitter, even a math result. "Let's wait a day or two for the inevitable expert writeup."
Benchmarks & Local AI
swyx on the RLM paper: benchmark gaming via "Temu T-bench"
swyx highlights a trajectory-comparison writeup buried in the RLM paper from Alex Zhang and Omar Khattab: an open secret of frontier training is that you don't need to train on the test set to goalseek a benchmark — training on test lookalikes works ("Temu T-bench"), and since open-weight releases almost never include the datasets/RL environments, there's plausible deniability. The paper explores applying NLP distance metrics to hidden trajectories as a detection approach. The quoted RLM finding is itself notable for harness-pilled readers: well-designed harnesses generalize for the transformer — RLMs trained only on short tasks fully generalize to 8–32× longer ones because the harness makes structurally similar tasks produce near-identical trajectories, even across domains (essay-author matching transferring to math-solution matching). Reply consensus: "held-out test set" is no longer enough for agent evals — you need provenance on the task generator.
Local AI notes
- Video: Ahmad Osman argues we'll get Kimi-K3-equivalent intelligence on a single RTX PRO 6000 within 18 months — "The Desktop Frontier," a watch-and-bookmark explainer of how the efficiency curve gets there. Boosted by swyx.
- Meanwhile, in the present: GLM-5.2 at home for $15,000 (via LLMJunky), and LLMJunky's periodic reminder of the pace of progress — what model 3D rendering looked like just six months ago.
Other Notes
- Armin Ronacher: "Never Enough." Prompted by two stories — the $550k/year couple where the husband handed off parenting to become his company's top AI user, and the founder who has Claude score her first dates for empathy — Ronacher wrote a short essay about Silicon Valley's new crazy: not rebellion, but fear of falling behind. "Every 'saved' hour is returned to a race with no finish line... Maybe we need the misfits who know when to stop." His related quip: in more than one way AI is giving crypto vibes lately.
- Silent speech interface. Adam Cohen Hillel demos talking to AI with mouth movements only — no sound, from your existing phone/computer (demo at 0:13 and 1:09). Boosted by swyx.
- Ronacher's other experiment: he asked a model for a Fil-C vs Rust demo of the different memory-model tradeoffs, then couldn't resist: "Men would rather adopt garbage collection like constructs in a system level language than use Rust."
- Steve Yegge is back at AI Engineer with, per swyx, a warning about vibecoding "in a post Gas Town world" that "u guys aren't ready for".