Open source6 min read
Reflection AI released Beam on October 5 — a 501B-parameter MoE with 23B active, pretrained on 23.8T tokens, with a 1M context window. Per Reflection's own benchmarks: Terminal Bench v2.1 at 80.1% (fifth on the September 2026 leaderboard), SWE-bench Verified at 80.9%, GPQA Diamond at 90.5%. Apache 2.0 weights are coming later in October. No date confirmed yet.
Open source5 min read
Mistral released Large 4 on October 6 — 1.05T total parameters, 49B active, natively multimodal, API-only for now. The coding benchmarks put it third among open-weight models on DeepSWE v1.1 at 61.7%, behind Kimi K3 and DeepSeek V4.1 Flash. Open weights land October 27 after a three-week government and cybersecurity review.
Open source5 min read
Tencent open-sourced Hy4 Preview on August 28: 770B total parameters, 49B active per token inference, 1M context window, Apache 2.0 license, $0.834/M input on OpenRouter. The model is real and the benchmarks are competitive. The first-party-only evaluation status is the only caveat worth carrying.
Open source5 min read
Z.ai released GLM-5.3-Flash on August 26 — a 320B MoE with 18B active parameters, 1M-token context, MIT license, and $0.075/M input at a 50% launch discount. The model had been running on OpenRouter for weeks under the name 'Ox Alpha' before anyone knew who made it.
Open source5 min read
Z.ai launched GLM-5.3 on August 14 with the highest CyberGym vulnerability-discovery score of any publicly evaluated model — 84.5%, ahead of Mythos 5 and GPT-5.6 Sol. The open weights are on hold for roughly two weeks. It's the right call, for reasons that go past liability.
Open source7 min read
Meta open-sourced Muse Glimmer on August 10 — a 30-billion-parameter model that downloads to your Mac or PC and runs entirely offline. No subscription, no data leaving your device. It's not the most capable AI available, but it's yours in a way cloud AI is not.
Open source8 min read
Alibaba released Qwen3.8-27B weights on August 14 under Apache 2.0. At 4-bit quantization, it runs in roughly 17GB of VRAM. Qwen's model card claims 89.2% on GPQA Diamond and 90.3% on LiveCodeBench v6 — numbers that would put a 27B dense model close to frontier closed-weight performance. DeepSWE 1.1 tripled from 13.3% to 42.2% in one version. Several benchmarks are first-party Alibaba evaluations; the ones that matter for verification are LiveCodeBench and GPQA Diamond, which have independent infrastructure.
Open source8 min read
DeepSeek put V4 Pro 0813 into general availability on August 12 with no blog post — 1.6T MoE, MIT open weights, $0.87/M output. The next day it launched Harness v0.1, an open-source Claude Code rival that pulled 33,000 GitHub stars in hours. That one got an announcement. Prices go up August 16.
Open source6 min read
Nvidia released Nemotron 3.5 Lightning on August 11 — a 30B-parameter open-weight MoE that activates only 3B parameters per inference step. It benchmarks near gpt-oss-120b at roughly a quarter the total parameters, runs on a single H100 at 256K context, and ships under OpenMDW-1.1 with open weights, training data, and recipes.
Open source6 min read
Kimi K3's open weights drop July 27 — 2.8 trillion parameters, the largest open-weight release ever. The model was compelling before OSTP accused Moonshot of distilling Fable 5 and Treasury threatened sanctions. The accusation changes the compliance picture for builders planning production deployment.
Open source6 min read
Moonshot AI released Kimi K3 on July 16 — 2.8 trillion total parameters, the largest open-weight model ever announced. API live now; weights on HuggingFace by July 27. On Program Bench and SWE Marathon it leads GPT-5.6 Sol and Claude Fable 5. How a Beijing startup trained a 2.8T model without disclosed access to H100s is the question the release notes don't answer.
Open source5 min read
Mira Murati's Thinking Machines Lab released its first model on July 15: Inkling, a 975-billion-parameter open-weight multimodal model under Apache 2.0. The most striking line in the announcement is what they said they didn't build: the strongest model available. Here's why that might be a more defensible position than it sounds.
Open source6 min read
PrismML released Bonsai 27B on July 14 — a 27-billion-parameter reasoning model compressed to 3.9GB via 1-bit quantization that runs on an iPhone 17 Pro at 11 tokens per second. Based on Qwen3.6 27B, Apache 2.0 licensed, retaining 90% of full-precision quality across 15 benchmarks. The on-device AI story has been building toward this size for two years.
Open source7 min read
Meituan confirmed on June 30 that 'Owl Alpha' — the anonymous model ranking first on OpenRouter's Hermes Agent framework for two months — is LongCat-2.0, a 1.6-trillion-parameter coding model trained entirely on Chinese domestic chips under an MIT license. The weights haven't shipped to Hugging Face yet, but the API is live.
Open source7 min read
Zhipu AI released GLM-5.2 on June 16, 2026 — a ~744B open-weight model under MIT license with a vendor-reported 62.1% on SWE-bench Pro. It arrived two days after the US government export-controlled Anthropic's two most capable models offline. The open-weight coding frontier just moved. The trust question is where it moved to.
Open source6 min read
Google released Gemma 4 12B on June 3, 2026 — a free, open-weight model that runs on a 16GB laptop and understands text, images, audio, and video under Apache 2.0. On GPQA Diamond it scores 78.8%, beating Google's own Gemma 3 27B. The practical implication: capable AI that runs entirely on your own device, offline, with no subscription and no data leaving your machine.
Open source6 min read
Between April 7 and April 24, Z.ai, MiniMax, Moonshot AI, and DeepSeek each shipped an open-weight frontier coding model. DeepSeek V4 Pro now scores 80.6% on SWE-bench Verified at $0.44/M input tokens — eleven times cheaper than Claude Opus 4.7. The benchmark parity story is old news. The cost math is what's changed.
Open source5 min read
Mistral released Medium 3.5 on April 29 — 128B dense, 77.6% on SWE-bench Verified, 256K context, $1.50/$7.50 per million tokens. Devstral 2 and Magistral are both retired into this one checkpoint. Here's what that means if you're building on the Mistral stack.