←
Reflection110 posts
Reflection avatar✓

Reflection ✓

@reflection_ai

Make intelligence open and accessible to all

◷ Joined June 2024▧ 26 Photos and videos
Reflection avatar
Reflection@reflection_ai·Oct 16

Beam is now available across platforms, with open weights under the Apache 2.0 license - in collaboration with an ecosystem of 40+ partners. We're releasing: - Weights, tokenizer, and reference implementation - BF16, FP8 and NVFP4 weights and a DFlash2 drafter - The open source harness we built and trained Beam on: Mirror CLI Weights and model card on Hugging Face: huggingface.co/reflection

Post attachment
Reflection avatar
Reflection@reflection_ai·19h

As advanced cyber capabilities spread across the industry, our objective is to give defenders access to tools that decrease marginal risk in the operating environment. Building Laser and the Trusted Access Program is a step toward that goal.

Reflection avatar
Reflection@reflection_ai·19h

Laser gives organizations a scalable way to adversarially test the security of their applications and networks. Findings are verified by attack graphs and captured evidence, so teams can take direct action and improve their systems. Laser is now available as a research preview to teams selected for the program.

1:14
Reflection avatar
Reflection@reflection_ai·19h

We’re introducing Reflection’s Trusted Access Program, to give vetted organizations access to advanced AI capabilities and tools for cybersecurity. We’re starting with Laser, an autonomous penetration-testing harness for Mirror, Reflection’s terminal agent. reflection.ai/blog/launching…

Reflection avatar
Reflection@reflection_ai·Nov 16

Beam built a fully interactive flood and sea-level rise simulator from a written spec to a working, testable tool. It was built entirely through Mirror, our open source harness. Beam handled the mapping, the bathtub flood model, neighborhood-level impact tracking, and water source placement.

Reflection avatar
Reflection@reflection_ai·Nov 5

We also developed the Open Safety Framework as a comprehensive methodology that covers systemic risks, refusal behavior, and adversarial robustness. This ensures safe and responsible releases at scale. reflection.ai/safety

Reflection avatar
Reflection@reflection_ai·Nov 5

We developed the Iris Safety Evaluations (ISE) to test whether Beam’s responses align with and uphold our safety policy in realistic contexts, such as in adversarial and agentic settings. Evaluated across cyber offense, chemical, biological, and radiological/nuclear categories, Beam holds higher refusal rates than leading Western and Chinese open models.

Post attachment
Reflection avatar
Reflection@reflection_ai·Nov 5

Beam achieves the highest refusal/pivot rate among the Western and Chinese open-weight models tested, declining 89% of harmful requests while successfully responding to 82% of safety-related harmless requests—demonstrating that open models can achieve strong safety and helpfulness without compromise.

Post attachment
Reflection avatar
Reflection@reflection_ai·Nov 5

Sharing more about how we trained and evaluated Beam, our first model, for safety: reflection.ai/blog/open-mode… In the coming weeks, we'll release additional safety research, evaluations, and tools under Project Iris, Reflection's open AI safety initiative. We are committed to prioritizing safety in every model we build—and sharing what we develop and learn so that researchers, developers, policymakers, and enterprises can build on our work, advancing safety across the broader AI ecosystem.

You.com avatar
⟳ Reflection retweeted
You.com@youdotcom·Nov 3

Beam by @reflection_ai is now available on Inference Router. It's a US-built open-weight model for coding and agentic tasks. Teams can run it in production today, with ZDR available on every call. you.com/router

Post attachment
Reflection avatar
Reflection@reflection_ai·Oct 30

In this demo, Beam built a real-time weather satellite tracker with Mirror, our open source harness, using publicly available orbital data. It fetches live positions, tracks coverage, and renders orbits on a 3D globe.

Ben Burtenshaw avatar
⟳ Reflection retweeted
Ben Burtenshaw@ben_burtenshaw·Oct 27

You can win $20K in cash prizes by building a demo, training a model, or creating a dataset with Beam by @reflection_ai! Prizes are split across 3 winners that we'll choose after the challenge ends on Monday, November 2nd. Plus, we're giving $25 hub credits to the first 400 to take part, just find the link in the page below. Rules: Build something cool with beam and drop it on the discussion! Anything that might be helpful to the world or a cool demo is allowed: - You could make a demo that uses the Beam model on Spaces - You could quantize it for your favorite hardware - You could generate a synthetic dataset for your favorite problem - You could build an RL environment with or for Beam - You could benchmark Beam and share the results. - You could surprise us! If you build something with Beam, then share it on the Hub to enter the challenge. We’ll collect all the links, select the best, and reach out to winners. huggingface.co/reflection/Be…

Post attachment
Reflection avatar
Reflection@reflection_ai·Oct 27

We ran ~1.3 billion sandboxes in our largest RL run for Beam - spanning RL training, evaluation, and ablations. With evals and other RL experiments running, the fleet peaked at 883,006 vCPUs (83% of allocatable) across 23 clusters on @googlecloud and @awscloud. We plan to use OpenEnv (now part of @AgenticAIFdn) for future runs, bringing our learnings to the open ecosystem.