Reflection AI unveils Beam, its first open-weight model, aimed at Chinese rivals
Nvidia-backed Reflection AI unveiled Beam, a 501-billion-parameter open-weight model it says matches leading Chinese open models on reasoning at lower inference cost. The performance claims have not been independently verified, and weights are due this month.
1 / 2
The story, neutrally told
Centre · 2Nvidia-backed Reflection AI unveiled Beam, its first open-weight AI model, on Monday 5 October 2026. TechCrunchN “Reflection AI is officially unveiling Beam, its first frontier, open-weight AI model.” Read at TechCrunch ↗ CNAN “Nvidia-backed Reflection AI debuted its first open-weight model named Beam on Monday” Read at CNA ↗ Centre · 2Reflection describes Beam as a text-only mixture-of-experts model with 501 billion total parameters, of which 23 billion are active for each task, pre-trained on 23.8 trillion tokens with a 1 million token context window. TechCrunchN “Beam is a 501-billion-parameter model with 23 billion active parameters. It was pre-trained on 23.8 trillion tokens and has a 1 million token context window.” Read at TechCrunch ↗ CNAN “Beam contains 501 billion total parameters” Read at CNA ↗ Centre · 2By comparison, Z.ai's GLM-5.2 has about 744 billion total parameters with 40 billion active. TechCrunchN “Z.ai’s GLM 5.2 has roughly 744 billion total parameters with 40 billion active.” Read at TechCrunch ↗ CNAN “Z.ai's GLM-5.2 has about 744 billion total parameters with 40 billion active parameters.” Read at CNA ↗
Centre · 1Reflection says Beam scores on par with GLM-5.2 on advanced reasoning benchmarks and outperforms today's leading Western open models while using "3-4x less inference compute"; TechCrunch notes these claims have not been independently verified. TechCrunchN “Reflection’s performance claims haven’t been independently verified, but on advanced reasoning benchmarks, Reflection says Beam scores on par with Z.ai’s GLM-5.2” Read at TechCrunch ↗ Centre · 1CNA reports Reflection saying Beam is competitive with GLM-5.2 and closing in on Qwen3.8-Max on coding and agentic tasks. CNAN “Reflection said Beam is competitive with Chinese AI startup Z.ai's GLM‑5.2 and is closing in on Qwen3.8‑Max on coding and agentic tasks.” Read at CNA ↗ Centre · 1Reflection calls Beam a "workhorse model" for enterprises, the public sector and developers, and says it was trained with high-compute reinforcement learning for reasoning, coding and agentic tasks. TechCrunchN “Reflection calls it a “workhorse model” for enterprises, the public sector, and developers.” Read at TechCrunch ↗
Centre · 1Reflection's own benchmarks show Beam outscoring Inkling, the open model from Thinking Machines Lab released in July, on four coding tests, though Inkling is multimodal and Beam is text-only. TechCrunchN “Reflection’s own benchmarks show that Beam outscores Inkling on four coding tests where both report results, but Inkling is a multimodal model and Beam is text-only.” Read at TechCrunch ↗ Centre · 1Reflection says it will release Beam's weights and full technical details this month, with distribution through hyperscalers and neoclouds and integrations across open source libraries at launch. TechCrunchN “Reflection says it will release Beam’s weights and full technical details this month” Read at TechCrunch ↗ Centre · 2Reflection was founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou, and has raised roughly $4.7 billion from backers including Nvidia, Sequoia Capital and Lightspeed Venture Partners, per PitchBook, at a $25 billion pre-money valuation in its last round. TechCrunchN “has raised roughly $4.7 billion from backers including Nvidia, Sequoia Capital, and Lightspeed Venture Partners, per PitchBook.” Read at TechCrunch ↗ CNAN “Founded in 2024 by former DeepMind researchers Misha Laskin and Ioannis Antonoglou” Read at CNA ↗
Centre · 2This summer Reflection signed deals collectively worth more than $7 billion with SpaceX and Nebius for access to Nvidia GB300 chips through 2029; CNA specifies the SpaceX deal covers capacity at Colossus 2. TechCrunchN “Reflection signed deals collectively worth more than $7 billion with SpaceX and Nebius to secure access to Nvidia’s GB300 chips through 2029.” Read at TechCrunch ↗ CNAN “Reflection signed a deal with SpaceX for additional computing capacity at the Elon Musk-led company's Colossus 2 data center.” Read at CNA ↗ Centre · 1Reflection is targeting enterprises and sovereign nations with "AI factories", letting institutions train its models on their own data, and has begun testing a sovereign AI factory partnership with Shinsegae Group in South Korea; TechCrunch notes the idea is championed by Nvidia CEO Jensen Huang and would benefit Nvidia's GPUs. TechCrunchN “Reflection has already begun testing the concept of a sovereign AI factory partnership with Shinsegae Group in South Korea.”“Nvidia CEO Jensen Huang, whose company backs Reflection, has long championed the “AI factory” idea” Read at TechCrunch ↗ Centre · 1CNA frames the launch as US tech firms trying to fend off cheaper, more customizable Chinese open-weight models that generate code nearly as well as leading OpenAI and Anthropic models. CNAN “US tech firms are trying to fend off competition from Chinese open-weight models that are cheaper, more customizable and capable of generating code nearly as well as leading models from OpenAI and Anthropic.” Read at CNA ↗
Centre · 1Reflection did not respond in time to TechCrunch's requests for more information. TechCrunchN “Reflection did not respond in time to TechCrunch’s requests for more information.” Read at TechCrunch ↗
Every sentence links to the reporting it rests on. The pill in front of each says where its sources sit: Left, Centre or Right when one side supplies at least half of them, Mixed when they are evenly split. The number is how many outlets it cites.
Left0 outlets
No left outlet in our sources has covered this story yet.
Centre2 outlets
- Framing
- Both outlets present Beam as a Western open-weight challenger to cheaper Chinese models; TechCrunch adds funding, compute and business-strategy detail and CNA gives a short, wire-style summary.
- Emphasis
- Parameter counts versus GLM-5.2, Reflection's benchmark claims, the race against Chinese open models, Nvidia backing and the SpaceX compute deal.
- Leaves out or plays down
- CNA omits that the claims are unverified, the weights release date, the Inkling comparison and the AI factory pitch; TechCrunch does not mention the Qwen3.8-Max comparison.
- Charged language
- “frontier”“workhorse model”“take on Chinese open models”
- For example
-
“a claim that could heat up the race to build a Western answer to DeepSeek, Qwen, and Z.ai.” — TechCrunch
“as the startup aims to compete with lower-cost Chinese models such as DeepSeek and Kimi in coding and agentic tasks.” — CNA
Right0 outlets
No right outlet in our sources has covered this story yet.
What every side reports
- Reflection AI unveiled Beam, its first open-weight model, on 5 October 2026.
- Beam has 501 billion total parameters and 23 billion active; GLM-5.2 has about 744 billion total and 40 billion active.
- Reflection says Beam is competitive with Z.ai's GLM-5.2.
- Reflection is backed by Nvidia and has a compute deal with SpaceX.
Where accounts differ
-
Whether Beam's performance claims hold
- Centre
- TechCrunch says Reflection's claims have not been independently verified; CNA relays them as the company's statements.
Reflection AI organisation
Says Beam matches leading Chinese open models on reasoning at 3-4x less inference compute, will publish weights this month, and is pitching AI factories to enterprises and sovereign nations.
“Reflection says Beam scores on par with Z.ai’s GLM-5.2 and outperforms today’s leading Western open models while using “3-4x less inference compute.”” — TechCrunch
Nvidia organisation
Investor in Reflection; its CEO Jensen Huang has championed the AI factory idea and a stronger open AI ecosystem, which TechCrunch notes would also benefit Nvidia's GPUs.
“Nvidia CEO Jensen Huang, whose company backs Reflection, has long championed the “AI factory” idea and pushed to strengthen the open AI ecosystem” — TechCrunch
North America1 outlet
TechCrunch focuses on Reflection's funding, compute deals and enterprise strategy, with caution on unverified benchmarks.
“Reflection’s performance claims haven’t been independently verified” — TechCrunch
TechCrunch
East and Southeast Asia1 outlet
CNA frames the launch as US firms countering cheaper Chinese open-weight models, with a short technical summary.
“The launch comes as US tech firms are trying to fend off competition from Chinese open-weight models” — CNA
CNA
Left0 articles
No coverage yet.
Centre2 articles
-
Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost
Neutral Detailed business-tech report that relays Reflection's claims while flagging they are unverified and noting Nvidia's interest.
-
CNAN ·
Nvidia-backed Reflection unveils first AI model to take on Chinese open models
Neutral Brief summary of the launch as a US challenge to cheaper Chinese open-weight models.
Right0 articles
No coverage yet.
- 5 Oct 20:33 First TechCrunchN Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost
- 5 Oct 21:49 +1h 16m CNAN Nvidia-backed Reflection unveils first AI model to take on Chinese open models
Times are when each article was published, or when we first saw it if the outlet gave no time.