Reflection

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:

RT @ben_burtenshaw: 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.

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.

Last week's special: free tokens. This week's special: free tokens. Try Beam here ↓

Lorenzo Porras (@thecodingsoup) built an escape room where Beam repairs a planetary machine, routes light through mirrors, and plans a robot's escape across bridges and doors. Beam receives the rules and state through tools, then writes and runs Python in @CoreWeave Sandbox to solve the puzzles. A 3D room shows the run as it happens.

One week since we dropped the weights for Beam. 10 trillion tokens served to date. Try the model for free:

RT @togethercompute: factory got early access to beam through together before launch they used that window to test @reflection_ai's model on real coding + agentic workflows, including tool-error recovery, conflict handling + final repo verification by day 0, @FactoryAI already knew how beam behaved on the work that matters to their customers

Adam @adamisnotroman made some throwback mobile browser snake games using Big Pickle and “Cat Jam" -- the code name for Beam during our stealth preview with OpenCode. The two models tackle the same prompt, with reported build times of 2:19.7 for Beam and 4:56.4 for Big Pickle.

We plugged Beam into @opencode, asked it to look at @UnslothAI, and create a fine-tuning notebook for the latest and smallest Gemma-4 model on a Text2SQL task – an out of distribution domain for Beam. The model successfully researches the model cards, Unsloth documentation, and available READMEs, and implements the fine-tuning logic and surrounding machinery. As a result, Beam is able to successfully finetune Gemma, increasing Gemma’s accuracy on its heldout test set by 66.5%.

RT @mercor: Mercor partnered with @Reflection_ai on Beam's training and evaluation. Beam scores 48.8% Pass@1 on APEX-Agents, #1 among US open-weight models and closely trailing Chinese open models such as GLM-5.3-Flash (52.8%). On APEX-Agents, which measures long-horizon, multistep tasks in professional services, Beam scores a 15-point lead among US open-weight models. Beam: 48.8% Muse Spark Open: 33.4% Nemotron 3.5 Super VL: 29.5% Nemotron 3 Ultra: 22.7% It trails Qwen 3.8-Max by 14.5 points and lands within the confidence interval of GLM-5.3-Flash, trailing by 4.0 points. Qwen 3.8-Max: 63.3% MiMo-V2.6-Flash RL: 57.4% GLM-5.3-Flash: 52.8% Beam: 48.8% Kimi K3: 47.8% Qwen 3.8-27B: 47.5% GLM-5.2: 45.2% Compared with GLM-5.3-Flash on APEX-Agents domains, it leads on management consulting by 1.5 points. It trails GLM-5.3-Flash by 11.8 points on corporate law, and investment banking by 1.6 points. Congratulations to the Reflection team. We’re excited for future collaboration together advancing the US open source frontier. See full leaderboard:

We teamed up with @huggingface to give away $20,000 to anyone who's building with Beam 🤗 Use Beam for any project - make a demo or a dataset, benchmark the model, create an RL environment - and comment your link in the model's repo page by Monday, November 2. We choose the winners. Winners choose cash or credits. Claim $25 in Hugging Face credits to build with Beam and enter the competition 👇

And if you run out of credits, get building on Tokens are still free.

Model customization recipes are live on our website and in the cookbook repo. Beam is now supported across RL and SFT stacks including @axolotl_ai, NeMo RL, NeMo AutoModel, Megatron Bridge (@NVIDIAAI), Miles (@radixark), SkyRL (@NovaSkyAI), and TRL (@huggingface). Get cooking 🧑‍🍳

RT @ValsAI: Reflection's Beam 1.0 now ranks #10 among all open-weight models, is the fastest model on the Vals Index, and is free to the public.

RT @ArtificialAnlys: US open-weights models made gains in both intelligence and parameter efficiency last week: ➤ Thursday morning: NVIDIA released Nemotron 3.5 Bolt and Nemotron 3.5 Super VL. Super VL scored 26.5 on the Intelligence Index with 120B parameters, pushing Inkling Small off the Pareto frontier ➤ Thursday midday: Meta's Muse Spark Open scored 38.9, more than 12 points ahead of Super VL and the highest intelligence yet for a US open weights model ➤ Friday morning: Beam 1.0, Reflection's first model, scored 30.3 with 501B parameters, joining the frontier between Super VL and Muse Spark Open See our full comparison of open weight models on Artificial Analysis

Cat's out of the bag - thanks to everyone who built with CatJam on @opencode. Beam is still free on OpenCode for the next week.

We recreated the viral X puzzle by asking Beam to create a fixed 180×90 grid for longitudes -179° to 179° and latitudes -89° to 89°, with 16,200 points. Despite not being trained on multimodal inputs like images, Beam runs the analysis in text mode and gets 95.5% coverage right, putting it between Opus 5 (92.5%) and Fable 5 (97.8%), which shows how well it generalizes to novel new tasks.

We asked Beam to create a live updating subway map for the NYC subway using the publicly available NYC MTA data. The model searches for documentation, checks authentication requirements, finds subway and map geometries, and gets creative moving the trains in between stations. Beam builds the right frontend and backend and ensures the server keeps running for live updates. [updated]