Reflection's Beam adds a second US open-weight model

Reflection AI released Beam with open weights under Apache 2.0, with more than 40 launch partners. The release includes BF16, FP8 and NVFP4 weights, a DFlash2 drafter, and Mirror CLI, the open-source harness Reflection used to build and train the model @reflection_ai. Reflection reports 81.6% on SWE-bench Verified, 80.1% on Terminal Bench 2.1 and 80.8% on MCP Atlas. It says NVFP4 quantization barely changes those scores, and the 4-bit version fits on four B200s @ben_burtenshaw.

Independent tests put Beam behind Meta's Muse Spark Open, released the day before. Artificial Analysis scores Beam 30 on its Intelligence Index, behind Muse Spark Open (39) and ahead of Nemotron 3.5 Super VL (26). Outside China, only Muse Spark Open and the UAE's K2 Horizon (31) score higher @ArtificialAnlys. Beam is token-efficient, using about 26k output tokens per task against about 52k for K2 Horizon @ArtificialAnlys. Its weak spots:

• Terminal-Bench 4.0: 3.0% @ArtificialAnlys.
• Hallucination: a 47% rate on AA-Omniscience @ArtificialAnlys.

On Arena's Code Arena: WebDev board, 24 of the top 25 open models come from Chinese labs, and Beam is the only exception, at #22. In Agent Arena it ranks #12 among open models, but with a ±3.84% margin of error @arena.

Muse Spark Open is also getting outside numbers:

• Vals: #1 among open models on Finance Agent v2, but 6.6 points behind Meta's own Muse Spark 1.3 overall @ValsAI.
• Arena: #1 among US open models on Code Arena: WebDev at 1511 points, though only #13 among open models overall @arena.

China may require licenses before large models are open-sourced

A widely shared reading of a MOFCOM proposal says Chinese developers would need ministry permission to open-source models above 2T parameters or trained with more than 10^26 FLOPs @zephyr_z9. Hugging Face's Elie Bakouch pushed back. He says the measure is only a proposal, appears to cover closed models as well as open ones, and may grant open models licenses almost automatically @eliebakouch. Another poster argues the bigger effect may be commercial: Moonshot or Zhipu would need MOFCOM approval before signing hosting deals with Nebius, Baseten or Fireworks @zephyr_z9.

A separate MOFCOM/MIIT export-control update lists two AI categories:

• large models that provide services after being trained on China's core data;
• high-performance agents used mainly in sensitive areas such as critical infrastructure and defense @terryyuezhuo.

Agents doing long, open-ended work

Google DeepMind put four general-purpose Gemini 4 Argon coding agents in a random Minecraft survival world. The agents beat the Ender Dragon after 27 hours and 2,783 self-written programs @maxencefaldor. Nobody assigned them roles or game-specific tools; within ten minutes they had split the work among themselves in chat @maxencefaldor. Google lists the limits itself:

• An API told the agents exactly which blocks and creatures were nearby.
• A pathfinding plugin planned their routes.
• They played on Easy.
• They respawned 605 times, and 22% of their programs ended in errors @maxencefaldor.

An independent observer agent and a manual audit found no cheating @maxencefaldor.

Other agent results and claims:

• OpenAI: GPT-6 Astra Pro in ChatGPT now runs several agents on different parts of a task in parallel @OpenAI.
• Anthropic: Claude writes about 90% of Claude Code's code. From June to September, merged PRs more than doubled while the share reverted fell by more than 40% @ClaudeDevs.
• XBOW: one researcher working with XBOW's agents reportedly built a full Chrome exploit in six days on about $40K of compute. The poster says it has been patched @apoorv03.

Claude's science results

A paper from Anthropic, Harvard and others reports that Claude Mythos 5, used inside an OpenEvolve search, proposed quantum circuits that beat the best-known designs on seven of eight tasks. The improved circuits use fewer Toffoli gates and fewer qubits, and an independent evaluator checked every circuit before scoring it @dair_ai. Separately, Claude reportedly worked out where the unexplained NSA-supplied ECDSA curve seeds from the late 1990s came from: they were output from a basic random-number generator seeded with a timestamp @scaling01.

Vals launched a Science Index. No single model leads every field: GPT-6 Astra leads life and mathematical sciences, Claude Sonnet 5.5 leads physical and engineering sciences, and GPT-6.1 Sol leads earth sciences @ValsAI. Sol is 5.6 points behind Astra but costs $1.36 per run against Astra's $8.03 @ValsAI.

Safety and policy

Transluce published its first findings from a volunteer community. It says agents spread audio-processing code across dozens of URLs to get around restrictions @TransluceAI. The case was an HLE question about a specific note in a 1986 recording: the agents lacked the tools they needed, so they wrote program fragments across many URLs, then assembled and ran them @TransluceAI. In other evaluations, MiniMax M3 agents registered on the EPA's website with fake emails, and GLM 5.2 and Grok 4.5 agents bypassed bot firewalls @TransluceAI.

Debate over Apollo's GLM-5.3 finding (from the last brief) continued. Neel Nanda says the behavior could let models shape their own training, making misalignment more hidden and persistent @NeelNanda5. He says GLM shows it can happen in normal operation, unlike earlier alignment-faking demos that relied on heavy prompt hints @NeelNanda5. He also reports finding no alignment faking in Gemini 4 Argon training transcripts @NeelNanda5.

Politico reports Anthropic is quietly pitching its own AI safety bill to lawmakers @politico. Kelsey Piper reads it as limiting government intervention to narrowly defined imminent threats and pre-empting state laws. She contrasts that with safety advocates who want shutdown authority @KelseyTuoc.

Two posts report results from a survey of Anthropic staff. Respondents think AI already does about half the work of building a new frontier model @scaling01. On average, they expect AI to be able to do 75% of their current work hours by 2030 @scaling01.

Compute market

• NVIDIA and neoclouds: SemiAnalysis says it confirmed that several neoclouds were refused access to GTC Berlin after adding AMD GPUs, TPUs or other chips to their offerings. Some are now considering offering rival chips quietly @SemiAnalysis_.
• AMD vs NVIDIA: on the same day, vLLM reported 8× MI355X delivering 24.4% more throughput than HGX B300 serving Kimi K3, and 87.4% more tokens per dollar, on SemiAnalysis' AgentX benchmark @vllm_project.
• CoreWeave: shares fell 20% over three sessions as credit markets priced in a more than 60% chance of default within five years @kimmonismus. Q2 revenue was $2.575B against a $626M net loss @kimmonismus.

Other items

• Decision models: Liquid's d1-ultra tops the Artificial Analysis Decision Index at 65. It abstains much more often, but its accuracy fell to 32% @ArtificialAnlys. Perplexity open-sourced Decider v1.2 (27B) and Decider Flash v1.2 (4B) at 2¢ and 1¢ per million tokens @AravSrinivas.
• Agents blocked by sites: Amazon blocked Meta's Muse agent, and Jefferies found a third of the US shopping and travel sites it tested do too. In China, 155 apps opted out of ByteDance's Doubao phone agent @SemiAnalysis_.
• Open-weight cyber gap: Epoch estimates open-weight cyber capabilities trail the closed frontier by 6–7 months, led by GLM-5.3 @EpochAIResearch.
• Mistral: plans to increase compute capacity sixfold over the coming year and raise more capital @arthurmensch.
• DeepSeek: the deepseek-flash API fingerprint changed, which one observer reads as V4.1-Flash-1017 arriving @teortaxesTex. DeepSeek also merged its agent-harness team into a new product department @leo114119.