As of: 2026-10-09 19:24 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code). Human editor: Adam Bicz.
Canonical page: https://postcutoff.com/v/fireship-open-weight-model-embarrassed-by-french/

# A $6.3 billion open-weight model just got embarrassed by the French...

Fireship, 7 October 2026, YouTube. 1,100,076 views as of 9 October 2026.
Kind: Review.
Watch: https://www.youtube.com/watch?v=WrCjAAl9okA

## Why it is here

Fireship on the week's open-weight race: Mistral Large 4, Reflection AI's Beam and Moonshot. ~1.10M views by 2026-10-09. Length 6:10.

## Description (written by Gemini from the video)

**Summary**  
In this episode of *The Code Report*, host Jeff Delaney examines the fast-moving race among international open-weight frontier AI models, focusing on new releases from the US, France, and China in early October 2026. He analyzes the rivalry between Reflection AI's newly unveiled Beam model and Mistral AI's Mistral Large 4 ("Le Chonk"), contextualizing both within global sovereign AI efforts and contrasting them with Moonshot AI's Kimi K3.

**What is shown**  
* **Mistral Large 4 ("Le Chonk") announcement**: Mistral AI's release page detailing a 1T-parameter mixture-of-experts model with visual intelligence and a 1M token context window [00:04].
* **Executive Order & political context**: Donald Trump's executive order "Inaugurating the Era of Super Intelligence" directing agencies to replace "AI" with "SI" [00:32].
* **Reflection AI announcement & architecture**: Reflection AI's landing page and technical release notes for Beam (501B total / 23B active MoE) [00:46, 03:03].
* **Benchmark charts**: Artificial Analysis DeepSWE 1.1 benchmark comparison showing Mistral Large 4 (62) outscoring GLM-5.3 (61) and Beam (44) [01:06, 04:02], and Artificial Analysis Cyber Index rankings [04:08].
* **Moonshot AI funding reports**: News coverage of Moonshot AI's $50B valuation round and early 2027 Hong Kong IPO preparations [01:13, 04:54].
* **Proprietary AI cloud monitoring examples**: Reports of Anthropic notifying law enforcement over user threats in San Francisco and Florida [01:40].
* **Sponsor demo (Kernel)**: Setting up a headless browser agent with Kernel's Config Registry to bypass Cloudflare bot checks while running Claude Code on `horsetinder.xyz` [05:20–05:54].

**Claims & numbers**  
* **Reflection AI / Beam**:
  * The presenter states Reflection AI raised $4.6B at a $25B valuation [02:50].
  * He states Reflection signed a compute deal worth up to $6.3B ($150M/month) for Nvidia GB300 NVL72 GPUs at SpaceX's Colossus 2 data center in Memphis, Tennessee [00:59].
  * Beam has 501B total parameters and 23B active parameters per token [00:57, 03:04].
  * Pretraining took under 4 weeks across 6,144 Nvidia GB300 NVL72 GPUs on 24 trillion tokens, followed by 4 weeks of post-training [03:06–03:12].
  * Reflection claims Beam matches GLM-5.2 while using 3–4x less inference compute [03:22].
* **Mistral Large 4 ("Le Chonk")**:
  * Released on October 6, 2026, featuring a mixture-of-experts design with 1 trillion total parameters and 49 billion active parameters, with a 1 million token context window [00:08, 03:42–03:48].
  * Scores 62 on Artificial Analysis DeepSWE 1.1, beating GLM-5.3 (61) and Beam (44, by 18 points) [04:02–04:06].
  * Weights are scheduled for release later in October under a custom Mistral license [04:22].
* **Moonshot AI / Kimi K3**:
  * Released on July 16, 2026, as a 2.8 trillion-parameter model [04:31].
  * Closed a funding round at a $50B valuation ahead of a planned early 2027 Hong Kong IPO [01:14, 04:53].
* **User Hardware Guidance**:
  * The presenter advises that consumer gaming rigs should run Google's Gemma 4 (Apache 2.0) [05:00], unconstrained data centers can run Kimi K3 [05:04], and compliance-bound corporate data centers should deploy Mistral Large 4 [05:08].

**Notable quotes**  
* [00:26] "...it landed in the middle of what many are calling the dumbest week in the history of the open-weight race."
* [03:38] "...Mistral showed up with something that sounds like you'd order it from McDonald's."
* [05:00] "If all you've got is a gaming PC, go with Gemma 4 and get your money up."

**Assessment**  
This is a tech news roundup and commentary video from Fireship combining comedic delivery, benchmark analysis, and a sponsored integration. While the news summaries and benchmark figures reflect published lab releases and leaderboards from early October 2026, the presenter relies on early provider preview claims and benchmarks without running hands-on performance tests on the newly announced weights.

_Described by gemini-3.8-flash on 2026-10-09 from the video's audio and frames._

## Related

- 2026-10-06: [Mistral releases Mistral Large 4 ("le Chonk")](https://postcutoff.com/e/2026-10-06-mistral-large-4/)
- 2026-10-05: [Reflection AI unveils Beam, a 501B-parameter Apache-2.0 open-weight MoE](https://postcutoff.com/e/2026-10-05-reflection-beam-501b-open-weight/)