Dagens Vibes — 22. september 2026

Fra X-feedet

Grok 4.7: hurtigt er ikke det samme som billigt

SpaceXAI lover et større grundlag, 500K context og bedre langvarigt arbejde til samme pris som 4.6: $2/$6 per million tokens. Ben Davis’ første vibe check er klart positivt på fart og placerer modellen over Sol og Opus til hverdagsarbejde — men stadig under Astra/Fable-klassen.

Ben Davis@davis7

To me so far the big news with Grok 4.7 is it's speed. The new "fast" mode is actually extremely fast (idk what the TPS is but it's really good) It's very obviously not Astra/Fable class, SpaceXAi (or anyone else really) has yet to make one so far. Grok 4.7 is pretty clearly competing with Opus/Sol, and I would definitely take Grok 4.7 over either of them Price is very good, my gut says it's more efficient but I don't have hard numbers on that yet, and it's got all the main capabilities u need to actually be useful these days: subagents, long running tasks, looping, etc. Supposedly they trained it on the Grok Bot harness so it's really good at conversation + knowledge work. Really curious how that plays out, good product is SpaceXAi's biggest advantage right now so getting the models tailor made for those could work. Especially if that lets them keep the model on the cheaper side so they can serve more of it

♥ 396 · ↻ 8 · 💬 27 · 🔖 38

x.com/davis7/status/2102096402452988364

Artificial Analysis måler samtidig 81.000 outputtokens per opgave mod 27.000 for Astra. 4.7 løfter coding-agent-resultatet markant, men generel intelligens flytter kun to point. Billige tokens kan stadig blive en dyr opgave — en klassiker fra cloud-regningen.

Morgan@morganlinton

Huge congrats to the whole team at @SpaceXAI, amazing results on the Intelligence Index for 4.7 🔥 Coding performance ahead of Sol now 👀

Artificial Analysis@ArtificialAnlys

Grok 4.7 scores 46 on the Artificial Analysis Intelligence Index to bring SpaceXAI into the top 4 AI labs. Coding Agent Index performance has also improved, overtaking GPT-5.6 Sol Grok 4.7 scores +2 points over Grok 4.6 on the Intelligence Index, with strong performance on agentic knowledge work tasks. We evaluated the new model at xhigh reasoning effort. Key takeaways: ➤ Grok 4.7 gains +111 Elo over Grok 4.6 (high) on AA-Briefcase, scoring 1657 Elo and placing it alongside Claude Opus 5 and Claude Fable 5.1 at the frontier. ➤ Grok 4.7 (xhigh) with Grok Build scores 56 on the Coding Agent Index, up +9 points from Grok 4.6 (xhigh), ranking 4th behind Claude Fable 5.1, GPT-6 Astra and Claude Opus 5. ➤ Outside agentic knowledge work, Grok 4.7 broadly matches Grok 4.6 (high), with improvements on Terminal-Bench 4.0 and GDP.pdf and regressions on AA-LCR and AutomationBench-AA. ➤ Grok 4.7 (xhigh) uses approximately 81k output tokens per Intelligence Index task, compared with 36k for Grok 4.6 (high) and 27k for GPT-6 Astra (max).

♥ 29 · ↻ 0 · 💬 3 · citeret opslag: ♥ 1,7k

x.com/morganlinton/status/2102094260778770557

Læs Artificial Analysis’ fulde evaluering

OpenAI løser problemer hurtigere end institutionerne kan fordøje dem

OpenAI siger, at en intern model siden 28. august har løst mere end 100 åbne matematiske problemer. Det nye uafhængige udvalg skal vurdere betydning og publicering — men har udtrykkeligt ingen indflydelse på tempoet i selve forskningen. Et akademisk sikkerhedsnet, ikke en håndbremse.

OpenAI@OpenAI

We’re working with an independent advisory group of mathematicians to help OpenAI responsibly share advances in AI and mathematics. The group will advise on how we assess and communicate new mathematical results, uphold academic and professional standards, and build tools that support mathematical research and learning. Through this work, we want mathematicians to be at the center of shaping how AI supports mathematical understanding and how its benefits reach the wider community. https://openai.com/index/advisory-group-on-mathematics-and-ai/

♥ 8,2k · ↻ 887 · 💬 698 · 🔖 1,5k

x.com/OpenAI/status/2102093145051943229

TechCrunch om udvalgets begrænsede mandat

Agentværktøjerne får selvindsigt

OpenCode 2 kan skrive eller ændre plugins, config og MCP-forbindelser midt i en kørende session og bruge ændringen i samme tur. Tricket er ikke magisk selvevolution, men genafspillelige transformationer over tom state — dejligt kedelig arkitektur.

dax@thdxr

kit finally finished the blog post he covers how OpenCode 2.0 is able to modify everything about itself in real time even during running sessions btw post whatever easter eggs you're able to find https://anoma.ly/notes/opencode-reloaded/

♥ 724 · ↻ 36 · 💬 79 · 🔖 358

x.com/thdxr/status/2102206622000967955

Claude Code går den anden vej rundt og tester, om dine skills faktisk hjælper. Beskriv et godt resultat, lad eval-kommandoen skrive cases, og mål forbedringen. Endelig en knap mærket “bevis det”.

Addy Osmani@addyosmani

Reminder: Claude Code can test if a skill or plugin actually improves Claude's answers. > claude plugin eval init Tell Claude what a good result looks like in your plugin's folder and it writes the test cases for you. > run claude plugin eval .

♥ 1,5k · ↻ 95 · 💬 72 · 🔖 1,7k

x.com/addyosmani/status/2101921871092658655

Den smarte agent møder den dumme adgangskode

Muse hentede David Cramers skattedokumenter elegant — efter at have fået hans credentials. Det opsummerer consumer-agenter ret præcist: flowet føles som fremtiden, sikkerhedsmodellen som et delt Google Sheet.

David Cramer@zeeg

Gave Muse credentials to download my K-1s. A little off putting, but the flow is done really well. Not sure what the future of all this looks like that doesn’t completely give up security

♥ 29 · ↻ 0 · 💬 7 · 🔖 3

x.com/zeeg/status/2101889987663040647

Nyhedsbonus

Amazon har allerede valgt sin sikkerhedsmodel: Muse må slet ikke komme ind. Agenten mødes nu af en besked om, at automatiseret shopping bryder Amazons vilkår. Agentisk handel er åbenbart fremtiden, men ikke på konkurrentens grund.

Cloudflares Python Workers er blevet GA efter to års preview. FastAPI, Django, Flask, databasesockets og AI/MCP-biblioteker kan køre i Pyodide oven på WebAssembly — Python på kanten, med lidt JavaScript gemt under gulvbrædderne.