Anchit Gupta

Lead Eng @ xAI, ex Meta, Stanford, IIT Bombay

[Anchit Gupta @anzhit]

Grok 4.20 was the most complex training run yet. We optimized for the highest intelligence density at each step without trading off real world usability through a novel recipe. More improvements and evals to come as the beta matures 🚀

[Arena.ai @arena]

Grok 4.20 beta1 (single agent) debuts #1 on Search Arena, and #4 overall in Text Arena!

Highlights:
- #1 in Search, scoring 1226, leading GPT-5.2 and Gemini-3
- #4 in Text, scoring 1492 on par with Gemini 3.1 Pro

Congrats to the @xAI team and @elonmusk on this impressive milestone!

[Anchit Gupta @anzhit]

@Saaaang94 Congrats looks strong!

[Anchit Gupta @anzhit]

@guohao_li @SpaceXAI @SpaceX @elonmusk Cool!

[Anchit Gupta @anzhit]

@yunta_tsai @SpaceXAI Real world usefulness and adoption at frontier companies like Tesla is the main test :)

[Anchit Gupta @anzhit]

Grok 4.5 generalizes to really hard long-horizon benchmarks! Something that seemed far-fetched just a few months ago, but is exactly what grit and rapid iteration by an awesome team and resources at @SpaceXAI make possible.

Proud to have helped shape the RL recipes behind it: coding, ML, and long-horizon tool use, a lot more to come!

[Rishi Desai @rishi_desai2]

Grok 4.5 has taken the #1 spot on the SWE-Marathon leaderboard.

More analysis on reward hacking, token usage, and long-horizon behavior coming soon :)

[Anchit Gupta @anzhit]

@KyriectionZhang MCP performance looking 🌶️

[Anchit Gupta @anzhit]

@JaydenDavisNC @SpaceXAI Awesome! There are still rough edges so feedback is welcome

[Anchit Gupta @anzhit]

@tstorm It had to do well, the yaml's carried your name 😉

[Anchit Gupta @anzhit]

@avaitopiper GG but gimme some more elo and tokens 🐐

[Anchit Gupta @anzhit]

@beffjezos LFG!

[Anchit Gupta @anzhit]

@AndrewCurran_ Yes. Wait for a lot more independent benchmarks to come out

[Anchit Gupta @anzhit]

Grok 4.5 is live! We rebuilt our RL recipes form the ground up for frontier agentic intelligence and efficiency. Upcoming longer horizon benchmarks might surprise 😉

[SpaceXAI @SpaceXAI]

Announcing Grok 4.5, our first model trained specifically for coding and agents. It was trained with Cursor and offers frontier intelligence at leading speeds and cost efficiency.

https://t.co/i8HpU7w64k

[Anchit Gupta @anzhit]

@scaling01 The evals might surprise 😉

[Anchit Gupta @anzhit]

@elonmusk @SpaceXAI Potato > Spud

[Anchit Gupta @anzhit]

Exponentials are funny, one would think you get used to them seeing the progress over past few years. But then a new one comes along that's even more fascinating than the last

[Anchit Gupta @anzhit]

Very cool! SWE benchmarks finally catching up to the frontier models. Time to retire contaminated relics like SWE Bench Pro/Verified

[Henry Kiss Ehrenberg @henryehrenberg]

We expect agents to act like senior engineers, but most benchmarks still evaluate them like interns.

Excited to introduce Senior SWE-Bench, an open-source and @harborframework-native benchmark that assesses agents as senior engineers on long-horizon tasks with realistically under-specified instructions.

We expect agents to build real features going on just a quick Slack message, nothing like the super technical instructions most benchmarks provide. Senior SWE-Bench fixes that.

Claude Opus 4.8 is the current leader at 24% high quality solves, but it took 117K tokens on average to get there. Claude Sonnet 5 looked like it was going to swoop in for the top spot, but we found it cheated on 26% of trials.

[Anchit Gupta @anzhit]

🧑‍🍳

[Elon Musk @elonmusk]

Grok 4.5, based on our 1.5T V9 foundation model, with Cursor data added in supplemental training, is now in private beta at SpaceX & Tesla. Early evals show performance close to, perhaps exceeding Opus.

RL is continuing to significantly improve the model, and the Grok Build harness gets better every day.

Nice work by all those involved!

Completely trained from scratch new models will be released by @SpaceX every month this year.

[Anchit Gupta @anzhit]

@jaixbhatia So much dropshipped slop going around in circles from benchmarks to models and back

[Anchit Gupta @anzhit]

@Abhilashaag1 @grok @imagine A much stronger model coming v soon! 😉

[Anchit Gupta @anzhit]

@agarwl_ Both together can work really well. Of-course the exact balance depends on how off-policy your student is from the teacher