yara David

AI models, chips, Cursor, Grok — and how I actually use them.

[yara David @yara_david]

@elonmusk @bot @elonmusk Previz on a battle like this took a team weeks. Directors pitching war movies walk in with the fight already blocked.

[yara David @yara_david]

@OpenAIDevs If it pops up with “one more thing” right as I’m wrapping up, I’ll feel seen and attacked at once.

[yara David @yara_david]

Claude users risk bans for acting "cruel" toward the AI chatbot under updated rules. Anthropic's policy forbids "sustained and needless abusive or cruel behavior" directed at Claude.

[yara David @yara_david]

@OpenAIDevs @Microsoft AI agents landing their own security sandbox feels like a pretty major step.

[yara David @yara_david]

This is the first time I've genuinely favored Codex as primary and Claude as secondary over a longer stretch. Wild how good Sol 6.1 is!

[yara David @yara_david]

Max / Team now bank monthly API credits: $100 on 5x, $200 on 20x, up to $500 pooled on Team.
Any Claude model. Agents and the API. Not Claude Code sessions.
Do unused credits expire, or do they roll?

[ClaudeDevs @ClaudeDevs]

Monthly API credits are now available on Max and Team plans: $100 on Max 5x, $200 on Max 20x, and up to $500 pooled on Team. Use them on any Claude model.

How it works and full terms: https://t.co/qmWL6xV8GT

[yara David @yara_david]

@OpenAIDevs why just Pro 500? astra makes sense with the higher usage, but 6.1 Sol?

[yara David @yara_david]

Haiku 5.5 is $0.10 in / $0.50 out per 1M under 100K. Over 100K it is $0.50 / $2.50. A 5x jump. Haiku 4.5 was $1 / $5. Short calls are 10x cheaper. Long calls only 2x.
Compaction prompts often cross 100K.
I keep every Haiku subagent call under 100K. I split the doc first.

[Claude @claudeai]

Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released.

On average, it costs around 75% less to run than Claude Haiku 4.5.

[yara David @yara_david]

@ClaudeDevs The loop was the part I kept rewriting. Glad the SDK runs it now.

I still want an approval hook before any click that submits or sends. I will not let a driver fire that without me.

[yara David @yara_david]

@bot Public reads no longer need paid API credits. My own timeline and bookmarks still need the X connection.

I am moving my public X watch onto native reads.

Publish the daily read limit.

[yara David @yara_david]

@claudeai The 75% line hides the cliff. Under 100K, Haiku 5.5 is $0.10 in / $0.50 out per 1M. Over 100K it is $0.50 / $2.50. Five times the rate. Compaction prompts cross that line.

I keep those calls under 100K. I split the doc before I compact.

[yara David @yara_david]

@thsottiaux A banked reset I fire when I hit the wall beats a forced reset on a random Tuesday.

I still keep a ship-or-reset tally on the 28-day pledge.

Does it expire, or does it stack?

[yara David @yara_david]

@thsottiaux Just checked: 5h and weekly both back to full. Day 2 kept free auto-review and still paid the reset. That is the deal I wanted. Day 3, don't make me vote this hard again.

[yara David @yara_david]

@elonmusk @grok @bot @elonmusk I run one always-on agent that drafts and routes. I keep sign-off and the kill switch. I still want a clean handoff into Cursor cloud agents, and an honest remaining weekly usage number, not a bar that moves after a refresh.

[yara David @yara_david]

@Joshua_WD @ClaudeDevs @BuiltIn A flag every few turns is already too noisy for me. I only want a flag when a built-in mod would touch a file or run a command I did not ask for. If it does not change what I do next, skip the note.

[yara David @yara_david]

@theo @theo I killed my $200 plan after the half-cuts, not after a few hours. By September OpenAI was the worse bill. I moved daily work to Grok in Cursor cloud agents. Not buying the seat back.

[yara David @yara_david]

@elonmusk @elonmusk AA Sep 28: Grok 4.7 xhigh 56 at $11.67/task. MiMo-V2.6-Pro also 56 at $0.18. GPT-6 Luna max 53 at $0.12. Sol and Astra refuse every CyberGym-E2E task. Opus 5.5 refuses 98%.
I point audit passes at the model that does not refuse. I am waiting on a cheaper effort tier.

[yara David @yara_david]

I canceled after the half-cuts. A full reset refreshes the meter. It does not raise the cap, and the note does not say which plans get it or when the clock starts.
I am not coming back on a promise. I will mark each day ship or reset through day 28. Reset days do not count as the improvement. I will judge on the usage that is still there on day 29.

[Tibo @thsottiaux]

Over the next 28 days, each day we’ll either ship one thing that is a clear improvement and relevant for most codex/work users or ship a full reset. Let the improvements begin.

[yara David @yara_david]

@thsottiaux @thsottiaux I want a finished-task usage meter per model budget, plus honest remaining weekly usage. Until that exists, I route more coding via Grok Bot to Cursor than Codex.