fastball

How is that relevant?

AFAIK it's just yolo mode, which doesn't actually do any checks like Claude Code's auto-mode (which has a model checking all commands for safety).

Are we sure an Astra hacker swarm didn't compromise arcprize.org's servers and exfiltrate the private eval set in order to achieve that 99%?

Codex's lack of auto-mode is what prevents me from using it for serious work compared to Claude Code.

Output tokens are 5x more expensive than input tokens, so I'm not sure "dominate" is entirely correct.

A conversation with 20 turns, 50k tok growth per turn, 1m tok context at end would price out like this:

Fable 5 ($1/M cache reads) ; cache reads 9.5M tok × $1.00 = $9.50 ; cache writes 1M tok × $12.50 = $12.50 ; output 1M tok × $50 = $50.00 ; total = $72.00

Fable 5.1 ($0.25/M cache reads) ; cache reads 9.5M tok × $0.25 = $2.38 ; cache writes 1M tok × $12.50 = $12.50 ; output 1M tok × $50 = $50.00 ; total = $64.88

So yes, cheaper, but not massively.

Say more?

I wish model providers would stop committing chart crimes in their releases.

- if you're gonna order the rest of the bar chart by rank, order your model accordingly.

- if you're gonna highlight a winner in a table of benchmarks, don't highlight your entire model row in the table.

Etc etc

Do large development teams care about Firefox?

The problem is that neither you nor I have evidence that says they should be paying, either.

Nth order effects are powerful things. A single person could change the trajectory of Louisiana in a meaningful way.

As much incentive as the oil & gas industry receives?

hundreds of millions unpaid and zero investigative journalism to try to get at the truth, besides "liens filed therefore Elon Musk is a cheapskate".

You can test for test vacuity with mutation testing, using tools like Stryker (for JS). In my experience this significantly helps models to write tests that actually test for correctness rather than aligned assumptions.

So we don't want companies to buy books and scan them and do whatever they want afterwards, but we also don't want to allow piracy of digital copies (the $1.5B Anthropic settlement)? Bit of a rock and a hard place for them.

What Microsoft models?

It would take like 30s of vibe-coding (including writing the prompt) to add support for AGENTS.md to claude code.

Well the difference is that most home water filtration systems (e.g. Brita et al) are made out of plastic, so even if their filters do reduce microplastic levels, making that claim to consumers ("our plastic filters help reduce plastic in your water") might be a hard sell.

It's not really a unit of measurement but rather just what you'd call the container[1], which in turn is 5 gallon / 19L. But nobody says "My 19L water jug", they they just say "My water gallon".

Since most people in Indonesia get their drinking water from these jugs, stating this filter's output in "gallon" terms is just the easiest way to communicate potential savings etc (since people pay per jug, so they think in jug terms, not in any other unit of measure).

[1] https://m.media-amazon.com/images/I/61fpdmJYUuL.jpg

Yeah, in Indonesia (and other parts of the world), the 5-gallon jug you would stick in e.g. a water cooler is called a "water gallon".

There is a company[1] in Indonesia that actually makes very nice versions of this tech.

[1] https://www.terrawaterindonesia.com/

Moral posturing also doesn't equate to moral or ethical goodness.

That said, effectiveness actually is a frequent shortcut to goodness, in that effectiveness often leads to efficiency and efficiency is one of the best methods of indiscriminate uplifting of people in an otherwise unfair and imbalanced world.

In my experience your criticism of the writing is valid in all Claude 5 models, so I wonder if it is somehow related to the new tokenizer introduced in gen 5.

A 2s LLM call is pretty slow.

Is it one or the other? You either need to keep everything in separate PRs or you need to pretend that services are contiguous where deployment order doesn't matter?

Do you do any cross workspace/repo work, or are you mostly constrained to a single namespace?

But the model is open weight?

The author didn't say "we (all software engineers)", so the only person including you in his class is you.

He's had a private Twitter for a while.

Should've called it Safestral.

Also I do like Mistral's seemingly newer strategy of focusing on smaller, more fine-tuned models for various use-cases, presumably the result of their large MoE models not competing effectively with the frontier models.

Arguably the majority of codebases are not available on the public internet