Grok 4.5 TerminalBench: 39% → 81%
Grok 4.5 jumped from 39% to 81% on TerminalBench compared to Grok 4.3, representing a dramatic improvement in agentic coding capability.
Snapshot · God Mode Podcast
Grok 4.5 Matches Claude Opus 4.8 at a Tenth of the Cost | GPT-5.6 Sol Ships Under Government Review | The 3-Model AI Stack | OpenAI Retracts SWE-bench | Cloudflare x402 | EP21
Explore episode
Jul 9, 2026
Where this was said
The qualitative era: picking a model is taste now
At 13:15 · chapter starts 10:51
As benchmarks converge, Luca makes a case that resonates across the table: we've entered a qualitative era. He uses Claude for complex legal research — specifically, finding cross-jurisdictional loopholes he likens to hunting cybersecurity bugs — because Fable's depth is unmatched for that kind of pioneering work. But for quick website builds and SEO ranking tools, he reaches for Codex because Claude is simply too slow. Ben adds that Grok was missing from agentic coding conversations until this week, and that the Grok Build subscribers who bought at 60% off a few weeks ago made a surprisingly good deal. Rik notes that Grok 4.5's jump on Artificial Analysis benchmarks — from around 1,000 to 1,500 points, and 39% to 81% on TerminalBench [1] — Rik "Grok benchmark: 1,000 → 1,500 points: On Artificial Analysis benchmarks, Grok jumped from approximately 1,000 points (Grok 4.3) to 1,500 po…" 12:58 — is a serious leap, even if it's still nitpicked.
The Qualitative Era: Model Choice Is Now a Matter of Taste
The quantitative gap between frontier models is closing fast. Luca argues we've entered a qualitative era where model selection is personal — you use Fable for legal research, Codex for quick builds, and Grok when cost matters. There's no universally right answer.
Grok benchmark: 1,000 → 1,500 points
On Artificial Analysis benchmarks, Grok jumped from approximately 1,000 points (Grok 4.3) to 1,500 points (Grok 4.5), landing just behind Claude Opus and Sonnet.
Grok 4.5 banking benchmark: 12% → 32%
Grok 4.5 more than doubled its banking benchmark score from 12% to 32% compared to Grok 4.3, indicating significant gains on finance-domain tasks.
Grok 4.5 Trained on Cursor Data — Data Is the Moat
Grok 4.5's benchmark jump didn't come from architectural breakthroughs — it came from data. Once SpaceX AI plugged into Cursor's coding dataset, the model leapfrogged Composer 2.5. Ben's point lands hard: training data, not model design, is the real competitive moat.
Similar snapshots
20 downloads/day at launch
PropGPT averaged 20 downloads per day right after launching on the App Store through influencer marketing.
4 months of rebuilding with no marketing
Eyal and Yali shut down all marketing and spent 4 months completely rebuilding PropGPT from scratch.
Over 40,000 downloads total
PropGPT has accumulated over 40,000 total downloads since launch.
LLM costs only $20/month
PropGPT's large language model (AI) operating costs are just $20 per month, and the cost is continually falling.
Ad monetization for games
Ad-based monetization works well for game apps where users spend extended time in-session, as seen with Grid and Wordle.
Ads don't work for tool apps
Tool-focused apps like PuffCount are poor candidates for ad monetization because users don't stay in-session long enough.
Hard paywall is unskippable
A hard paywall is a screen that blocks all app features unless the user pays or starts a free trial — it cannot be dismissed.
Two core monetization models
Mobile apps are primarily monetized through either ads (best for games) or in-app purchases/subscriptions (best for tools).
YouTube beats every other platform
According to the episode, YouTube outperforms every other social platform for building trust and driving SaaS conversions.
