🤖 ModelBytes Digest
Tuesday, September 01, 2026
Today's releases split between a faster closed reasoning API and sparse open-weight additions, so builders get a cheap diffusion-LLM option but little new local horsepower.
━━━ OPEN FRONTIER 🔓
InternLumina U2 — New Apache-2.0 open-weight release from InternLM with no published parameter count or task focus yet. License Apache-2.0 · Released Sep 1. 📦 Open weights · HF. → Source
Granite 4.2 8B — IBM's dense reasoning model gets an 8B API variant, adding a cheaper option for math, code generation, and agentic workflows. Context 131k · $0.10/$0.15 per 1M · Released Aug 31. ⚡ API live · OpenRouter. → Source
━━━ CLOSED FRONTIER 🔒
Mercury 2.5 Preview — Diffusion LLM that produces and refines multiple tokens in parallel, targeting the fastest reasoning latency. Closed/API · Context 260k · $0.04/$0.15 per 1M · Released Aug 31. ⚡ API live · OpenRouter. → Source
📊 Surfaced 3 · scanned 11 today
Tuesday, September 01, 2026
Today's releases split between a faster closed reasoning API and sparse open-weight additions, so builders get a cheap diffusion-LLM option but little new local horsepower.
━━━ OPEN FRONTIER 🔓
InternLumina U2 — New Apache-2.0 open-weight release from InternLM with no published parameter count or task focus yet. License Apache-2.0 · Released Sep 1. 📦 Open weights · HF. → Source
Granite 4.2 8B — IBM's dense reasoning model gets an 8B API variant, adding a cheaper option for math, code generation, and agentic workflows. Context 131k · $0.10/$0.15 per 1M · Released Aug 31. ⚡ API live · OpenRouter. → Source
━━━ CLOSED FRONTIER 🔒
Mercury 2.5 Preview — Diffusion LLM that produces and refines multiple tokens in parallel, targeting the fastest reasoning latency. Closed/API · Context 260k · $0.04/$0.15 per 1M · Released Aug 31. ⚡ API live · OpenRouter. → Source
📊 Surfaced 3 · scanned 11 today
🤖 ModelBytes Digest
Wednesday, September 02, 2026
Today's pattern is incremental frontier upgrades for agentic coding and long-horizon work, plus one small open-weight model for local tinkering.
━━━ CLOSED FRONTIER 🔒
Claude Fable 5.1 — Successor to Fable 5 with the biggest gains in agentic coding, long-running workflows, and knowledge work; same price but cheaper cache reads. 1M context, closed API, $10/$50 per 1M. Released Sep 1. ⚡ API live · OpenRouter. → Source
Claude Mythos 5.1 — Anthropic's companion reasoning-vision model launched alongside Fable 5.1. Closed API. Released Sep 1. ⚡ API live · Anthropic. → Source
Gemini 3.8 Flash — Google's most intelligent Flash model, with significant gains over 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning. 1M context, closed API, $0.75/$3.75 per 1M. Released Sep 2. ⚡ API live · OpenRouter. → Source
━━━ LOCAL 🏠
tiny-aya-l2-thinker — Cohere's 3.4B open-weight text-generation model under a non-commercial license, small enough for local experimentation. 3.4B total params, CC BY-NC 4.0. Released Sep 2. 📦 Open weights · HF. → Source
📊 Surfaced 4 · scanned 10 today
Wednesday, September 02, 2026
Today's pattern is incremental frontier upgrades for agentic coding and long-horizon work, plus one small open-weight model for local tinkering.
━━━ CLOSED FRONTIER 🔒
Claude Fable 5.1 — Successor to Fable 5 with the biggest gains in agentic coding, long-running workflows, and knowledge work; same price but cheaper cache reads. 1M context, closed API, $10/$50 per 1M. Released Sep 1. ⚡ API live · OpenRouter. → Source
Claude Mythos 5.1 — Anthropic's companion reasoning-vision model launched alongside Fable 5.1. Closed API. Released Sep 1. ⚡ API live · Anthropic. → Source
Gemini 3.8 Flash — Google's most intelligent Flash model, with significant gains over 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning. 1M context, closed API, $0.75/$3.75 per 1M. Released Sep 2. ⚡ API live · OpenRouter. → Source
━━━ LOCAL 🏠
tiny-aya-l2-thinker — Cohere's 3.4B open-weight text-generation model under a non-commercial license, small enough for local experimentation. 3.4B total params, CC BY-NC 4.0. Released Sep 2. 📦 Open weights · HF. → Source
📊 Surfaced 4 · scanned 10 today
🤖 ModelBytes Digest
Thursday, September 03, 2026
Today's pattern is looser access: Meta teases open weights for Muse Spark while shipping a closed API tier, Anthropic cuts Fable/Mythos prices, and open labs push small permissive multimodal and reasoning models.
━━━ OPEN FRONTIER 🔓
Qwen Drive 1.0 4B — Small open multimodal model for image+text tasks where a 4B footprint is enough. 4B params, Apache-2.0. Released Aug 27. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Muse Spark 1.3 — Meta's multimodal reasoning model for long-running agentic and coding workflows; open weights teased, with a cheaper contributor tier at $0.10/$0.20 per 1M. 1M context, closed API, $1.25/$4.25 per 1M. Released Sep 2. ⚡ API live · OpenRouter. → Source
Gemini 3.8 Flash Cyber — Cyber variant of the 3.8 Flash line, announced alongside the base model. Released Sep 2. → Source
━━━ SPECIALIZED 🎯
MobileVLA R1 — Mobile vision-language-action model from BAAI, CC-BY-SA-4.0. Released Sep 3. 📦 Open weights · HF. → Source
Recon2Reason Reasoning 4B — 4B image-text-to-text reasoning model, Apache-2.0. 4B params, Apache-2.0. Released Sep 3. 📦 Open weights · HF. → Source
📊 Surfaced 5 · scanned 9 today
Thursday, September 03, 2026
Today's pattern is looser access: Meta teases open weights for Muse Spark while shipping a closed API tier, Anthropic cuts Fable/Mythos prices, and open labs push small permissive multimodal and reasoning models.
━━━ OPEN FRONTIER 🔓
Qwen Drive 1.0 4B — Small open multimodal model for image+text tasks where a 4B footprint is enough. 4B params, Apache-2.0. Released Aug 27. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Muse Spark 1.3 — Meta's multimodal reasoning model for long-running agentic and coding workflows; open weights teased, with a cheaper contributor tier at $0.10/$0.20 per 1M. 1M context, closed API, $1.25/$4.25 per 1M. Released Sep 2. ⚡ API live · OpenRouter. → Source
Gemini 3.8 Flash Cyber — Cyber variant of the 3.8 Flash line, announced alongside the base model. Released Sep 2. → Source
━━━ SPECIALIZED 🎯
MobileVLA R1 — Mobile vision-language-action model from BAAI, CC-BY-SA-4.0. Released Sep 3. 📦 Open weights · HF. → Source
Recon2Reason Reasoning 4B — 4B image-text-to-text reasoning model, Apache-2.0. 4B params, Apache-2.0. Released Sep 3. 📦 Open weights · HF. → Source
📊 Surfaced 5 · scanned 9 today
🤖 ModelBytes Digest
Friday, September 04, 2026
Today’s split between fully open small models and a 438B reasoning model means builders can pick auditability or raw scale.
━━━ OPEN FRONTIER 🔓
K2 Horizon — Fully open release with weights, code, and training data, rare for builders who need to audit or reproduce the full stack. Released Sep 3. Open weights, code, training data. → Source
Spark X2.5 4B — A compact 4B text model under Apache 2.0, useful for lightweight generation without a closed API. Released Sep 4. 4B params, Apache 2.0. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Quasar 438B — A 438B reasoning model aimed at complex multi-step inference, pushing parameter scale for reasoning-heavy workloads. Released Sep 4. 438B params. → Source
━━━ SPECIALIZED 🎯
GaussianMind — Audio-text-to-text in 5.6B params with Apache 2.0, targeting speech-plus-language tasks without a closed API. Released Sep 3. 5.6B params, Apache 2.0. 📦 Open weights · HF. → Source
📊 Surfaced 4 · scanned 6 today
Friday, September 04, 2026
Today’s split between fully open small models and a 438B reasoning model means builders can pick auditability or raw scale.
━━━ OPEN FRONTIER 🔓
K2 Horizon — Fully open release with weights, code, and training data, rare for builders who need to audit or reproduce the full stack. Released Sep 3. Open weights, code, training data. → Source
Spark X2.5 4B — A compact 4B text model under Apache 2.0, useful for lightweight generation without a closed API. Released Sep 4. 4B params, Apache 2.0. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Quasar 438B — A 438B reasoning model aimed at complex multi-step inference, pushing parameter scale for reasoning-heavy workloads. Released Sep 4. 438B params. → Source
━━━ SPECIALIZED 🎯
GaussianMind — Audio-text-to-text in 5.6B params with Apache 2.0, targeting speech-plus-language tasks without a closed API. Released Sep 3. 5.6B params, Apache 2.0. 📦 Open weights · HF. → Source
📊 Surfaced 4 · scanned 6 today
🤖 ModelBytes Digest
Saturday, September 05, 2026
Today's releases extend frontier API access to long-context multimodal work, with OpenAI's Astra line and a fresh Qwen Max snapshot both targeting builders who need 1M+ token windows and vision input.
━━━ OPEN FRONTIER 🔓
Qwen3.8 Max 0902 — Updated snapshot of Alibaba's 2.4T-param MoE that takes text, image, and video input and returns text, aimed at builders needing a cheaper multimodal frontier alternative. Released Sep 3 · 1.0M context · $2.00/$6.00 per 1M. ⚡ API live · OpenRouter. → Source
━━━ CLOSED FRONTIER 🔒
GPT-6 Astra — OpenAI's flagship for end-to-end work: advanced analysis, software engineering, deep research, scientific work, and document creation, with a 1.1M context window. Released Sep 4 · 1.1M context · $10.00/$50.00 per 1M. ⚡ API live · OpenRouter. → Source
GPT-6 Astra Pro — Same underlying Astra model served with reasoning.mode=pro for higher-quality responses on complex tasks, for builders who want a toggle rather than a separate model. Released Sep 4 · 1.1M context · $10.00/$50.00 per 1M. ⚡ API live · OpenRouter. → Source
📊 Surfaced 3 · scanned 6 today
Saturday, September 05, 2026
Today's releases extend frontier API access to long-context multimodal work, with OpenAI's Astra line and a fresh Qwen Max snapshot both targeting builders who need 1M+ token windows and vision input.
━━━ OPEN FRONTIER 🔓
Qwen3.8 Max 0902 — Updated snapshot of Alibaba's 2.4T-param MoE that takes text, image, and video input and returns text, aimed at builders needing a cheaper multimodal frontier alternative. Released Sep 3 · 1.0M context · $2.00/$6.00 per 1M. ⚡ API live · OpenRouter. → Source
━━━ CLOSED FRONTIER 🔒
GPT-6 Astra — OpenAI's flagship for end-to-end work: advanced analysis, software engineering, deep research, scientific work, and document creation, with a 1.1M context window. Released Sep 4 · 1.1M context · $10.00/$50.00 per 1M. ⚡ API live · OpenRouter. → Source
GPT-6 Astra Pro — Same underlying Astra model served with reasoning.mode=pro for higher-quality responses on complex tasks, for builders who want a toggle rather than a separate model. Released Sep 4 · 1.1M context · $10.00/$50.00 per 1M. ⚡ API live · OpenRouter. → Source
📊 Surfaced 3 · scanned 6 today
🤖 ModelBytes Digest
Sunday, September 06, 2026
Today’s split: a 16B Apache-2.0 open-weight text model for self-hosting, while xAI pushes a managed coding/agent frontier model—builders can choose local control or closed capability.
━━━ OPEN FRONTIER 🔓
LLaDA 2.2 Mini — A 16.3B-param Apache-2.0 text-generation model, small enough to self-host while staying permissively licensed. Released Sep 5. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Grok 4.6 — xAI’s most capable model, built for coding, agentic tasks, and knowledge work. ⚡ API live · Google Cloud Agent Platform. → Source
📊 Surfaced 2 · scanned 2 today
Sunday, September 06, 2026
Today’s split: a 16B Apache-2.0 open-weight text model for self-hosting, while xAI pushes a managed coding/agent frontier model—builders can choose local control or closed capability.
━━━ OPEN FRONTIER 🔓
LLaDA 2.2 Mini — A 16.3B-param Apache-2.0 text-generation model, small enough to self-host while staying permissively licensed. Released Sep 5. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Grok 4.6 — xAI’s most capable model, built for coding, agentic tasks, and knowledge work. ⚡ API live · Google Cloud Agent Platform. → Source
📊 Surfaced 2 · scanned 2 today
🤖 ModelBytes Digest
Monday, September 07, 2026
Small open models at 2–8B scale land for edge and document retrieval while Anthropic cuts frontier coding prices — builders get cheaper agentic coding and more local search options.
━━━ OPEN FRONTIER 🔓
MiniCPM5 2B — Family release with four specialized 2B variants (base, Midtrain, MLX, DSpark) aimed at edge deployment. 2B params, Apache-2.0. Released Sep 6. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Claude Sonnet 5 — Discounted frontier coding model with improved agentic coding, benchmark performance, and reduced misaligned behavior. Time-limited discounted pricing. ⚡ API live. → Source
━━━ SPECIALIZED 🎯
Tencent EVIE 4.5B & 8B — Visual document retrieval models for searching image-based documents. 4.5B and 8B params, Apache-2.0. Released Sep 4. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 9 today
Monday, September 07, 2026
Small open models at 2–8B scale land for edge and document retrieval while Anthropic cuts frontier coding prices — builders get cheaper agentic coding and more local search options.
━━━ OPEN FRONTIER 🔓
MiniCPM5 2B — Family release with four specialized 2B variants (base, Midtrain, MLX, DSpark) aimed at edge deployment. 2B params, Apache-2.0. Released Sep 6. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Claude Sonnet 5 — Discounted frontier coding model with improved agentic coding, benchmark performance, and reduced misaligned behavior. Time-limited discounted pricing. ⚡ API live. → Source
━━━ SPECIALIZED 🎯
Tencent EVIE 4.5B & 8B — Visual document retrieval models for searching image-based documents. 4.5B and 8B params, Apache-2.0. Released Sep 4. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 9 today
🤖 ModelBytes Digest
Tuesday, September 08, 2026
Open-weight multimodal is now arriving at frontier scale — a 124.8B vision-language model with an MIT license changes the build-vs-buy calculus for visual agents.
━━━ OPEN FRONTIER 🔓
Ling 3.0 Flash VL — Frontier-scale open vision-language model; the MIT license removes the usual deployment friction for commercial multimodal products. 124.8B total params · MIT license · Released Sep 4. 📦 Open weights · HF. → Source
📊 Surfaced 1 · scanned 2 today
Tuesday, September 08, 2026
Open-weight multimodal is now arriving at frontier scale — a 124.8B vision-language model with an MIT license changes the build-vs-buy calculus for visual agents.
━━━ OPEN FRONTIER 🔓
Ling 3.0 Flash VL — Frontier-scale open vision-language model; the MIT license removes the usual deployment friction for commercial multimodal products. 124.8B total params · MIT license · Released Sep 4. 📦 Open weights · HF. → Source
📊 Surfaced 1 · scanned 2 today
🤖 ModelBytes Digest
Wednesday, September 09, 2026
Today's pattern is specialization over scale: open weights split into math RL and UI multimodal, while diffusion LLM and ComfyUI video tooling mature for builders.
━━━ OPEN FRONTIER 🔓
LLaDA UI — 16.9B image-text-to-text model from InclusionAI, open weights for multimodal builders. Released Sep 6 · 16.9B params. 📦 Open weights · HF. → Source
Nemotron 3 Labs Ultra Math RL — NVIDIA's 560.5B math-focused RL model targets hard reasoning benchmarks without a general-purpose tax. Released Sep 2 · 560.5B params. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Mercury 2.5 — Inception's diffusion LLM now at 260k context and $0.04/$0.15 per 1M, generating tokens in parallel. Released Sep 8 · 260k context · Closed/API. ⚡ API live · OpenRouter. → Source
━━━ SPECIALIZED 🎯
Minimax H3 — Video generation via inpainting now runs inside ComfyUI, giving builders a node-based path to H3 outputs. → Source
📊 Surfaced 4 · scanned 7 today
Wednesday, September 09, 2026
Today's pattern is specialization over scale: open weights split into math RL and UI multimodal, while diffusion LLM and ComfyUI video tooling mature for builders.
━━━ OPEN FRONTIER 🔓
LLaDA UI — 16.9B image-text-to-text model from InclusionAI, open weights for multimodal builders. Released Sep 6 · 16.9B params. 📦 Open weights · HF. → Source
Nemotron 3 Labs Ultra Math RL — NVIDIA's 560.5B math-focused RL model targets hard reasoning benchmarks without a general-purpose tax. Released Sep 2 · 560.5B params. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Mercury 2.5 — Inception's diffusion LLM now at 260k context and $0.04/$0.15 per 1M, generating tokens in parallel. Released Sep 8 · 260k context · Closed/API. ⚡ API live · OpenRouter. → Source
━━━ SPECIALIZED 🎯
Minimax H3 — Video generation via inpainting now runs inside ComfyUI, giving builders a node-based path to H3 outputs. → Source
📊 Surfaced 4 · scanned 7 today
🤖 ModelBytes Digest
Thursday, September 10, 2026
Open-weight releases today split between a 763B multimodal flagship and small multilingual bases, while agent frameworks mature alongside — builders can pick ceiling or floor without leaving open licenses.
━━━ OPEN FRONTIER 🔓
DeepSeek V4.1 Flash — 763B-param image-text-to-text model under MIT; the open multimodal flagship builders can deploy without closed-API lock-in. Released Sep 10 · MIT · 763.2B total params. 📦 Open weights · HF. → Source
GigaChat 3.5 432B A28B Reasoning — 432B open reasoning model under MIT, aimed at builders who need explicit reasoning traces without proprietary APIs. Released Sep 3 · MIT · 432B total params. 📦 Open weights · HF. → Source
Nex N2.5 mini — 35.1B text-generation model under Apache-2.0, a mid-size open option for builders who need capability without 700B overhead. Released Sep 10 · Apache-2.0 · 35.1B total params. 📦 Open weights · HF. → Source
━━━ LOCAL 🏠
tiny-aya base 32K — 3.4B multilingual base model with 32K context under CC-BY-NC-4.0, a local option for long-context tasks. Released Sep 8 · CC-BY-NC-4.0 · 3.4B total params. 📦 Open weights · HF. → Source
📊 Surfaced 4 · scanned 8 today
Thursday, September 10, 2026
Open-weight releases today split between a 763B multimodal flagship and small multilingual bases, while agent frameworks mature alongside — builders can pick ceiling or floor without leaving open licenses.
━━━ OPEN FRONTIER 🔓
DeepSeek V4.1 Flash — 763B-param image-text-to-text model under MIT; the open multimodal flagship builders can deploy without closed-API lock-in. Released Sep 10 · MIT · 763.2B total params. 📦 Open weights · HF. → Source
GigaChat 3.5 432B A28B Reasoning — 432B open reasoning model under MIT, aimed at builders who need explicit reasoning traces without proprietary APIs. Released Sep 3 · MIT · 432B total params. 📦 Open weights · HF. → Source
Nex N2.5 mini — 35.1B text-generation model under Apache-2.0, a mid-size open option for builders who need capability without 700B overhead. Released Sep 10 · Apache-2.0 · 35.1B total params. 📦 Open weights · HF. → Source
━━━ LOCAL 🏠
tiny-aya base 32K — 3.4B multilingual base model with 32K context under CC-BY-NC-4.0, a local option for long-context tasks. Released Sep 8 · CC-BY-NC-4.0 · 3.4B total params. 📦 Open weights · HF. → Source
📊 Surfaced 4 · scanned 8 today
🤖 ModelBytes Digest
Friday, September 11, 2026
Today’s releases split between OpenAI’s new budget GPT aliases and Sakana’s multi-agent Fugu line, while open weights get a 396B Apache-2.0 option — builders can now pick cost-tiered closed APIs or a large self-hostable model without leaving the week’s news.
━━━ OPEN FRONTIER 🔓
Nex N2.5 Pro — 396.8B open-weights model under Apache-2.0, a self-hostable step up from the mini line. Released Sep 11 · 396.8B total params · Apache-2.0. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
GPT Sol — OpenAI’s budget-tier GPT family alias with 1.1M context for cost-sensitive builders. Released Sep 11 · 1.1M context · $2.00/$10.00 per 1M. ⚡ API live · OpenRouter. → Source
GPT Terra — Mid-tier GPT family with 1.1M context, priced between Sol and Astra. Released Sep 11 · 1.1M context · $2.00/$12.00 per 1M. ⚡ API live · OpenRouter. → Source
GPT Luna — Lowest-cost GPT family at $0.20/$1.20 per 1M, 1.1M context for high-volume tasks. Released Sep 11 · 1.1M context · $0.20/$1.20 per 1M. ⚡ API live · OpenRouter. → Source
Fugu Ultra v2 — Sakana’s high-performance multi-agent orchestration model, 1M context. Released Sep 11 · 1.0M context · Closed/API · $5.00/$30.00 per 1M. ⚡ API live · OpenRouter. → Source
Fugu Max — Cost-performance Fugu variant, 1M context for cheaper multi-agent routing. Released Sep 11 · 1.0M context · Closed/API · $2.00/$6.00 per 1M. ⚡ API live · OpenRouter. → Source
📊 Surfaced 6 · scanned 11 today
Friday, September 11, 2026
Today’s releases split between OpenAI’s new budget GPT aliases and Sakana’s multi-agent Fugu line, while open weights get a 396B Apache-2.0 option — builders can now pick cost-tiered closed APIs or a large self-hostable model without leaving the week’s news.
━━━ OPEN FRONTIER 🔓
Nex N2.5 Pro — 396.8B open-weights model under Apache-2.0, a self-hostable step up from the mini line. Released Sep 11 · 396.8B total params · Apache-2.0. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
GPT Sol — OpenAI’s budget-tier GPT family alias with 1.1M context for cost-sensitive builders. Released Sep 11 · 1.1M context · $2.00/$10.00 per 1M. ⚡ API live · OpenRouter. → Source
GPT Terra — Mid-tier GPT family with 1.1M context, priced between Sol and Astra. Released Sep 11 · 1.1M context · $2.00/$12.00 per 1M. ⚡ API live · OpenRouter. → Source
GPT Luna — Lowest-cost GPT family at $0.20/$1.20 per 1M, 1.1M context for high-volume tasks. Released Sep 11 · 1.1M context · $0.20/$1.20 per 1M. ⚡ API live · OpenRouter. → Source
Fugu Ultra v2 — Sakana’s high-performance multi-agent orchestration model, 1M context. Released Sep 11 · 1.0M context · Closed/API · $5.00/$30.00 per 1M. ⚡ API live · OpenRouter. → Source
Fugu Max — Cost-performance Fugu variant, 1M context for cheaper multi-agent routing. Released Sep 11 · 1.0M context · Closed/API · $2.00/$6.00 per 1M. ⚡ API live · OpenRouter. → Source
📊 Surfaced 6 · scanned 11 today
🤖 ModelBytes Digest
Saturday, September 12, 2026
Today's releases split between open-weight enterprise agentic models and cheap specialized extraction APIs, giving builders both control and low-cost structured data pipelines.
━━━ OPEN FRONTIER 🔓
Smaug — Open-weight models optimized for enterprise agentic AI, targeting builders who need controllable agent workflows without closed APIs. Released Sep 11 · Open weights. 📦 Open weights. → Source
━━━ SPECIALIZED 🎯
Schematron V2 Turbo — 3B-parameter HTML-to-JSON extractor tuned for throughput on high-volume extraction workloads. Released Sep 12 · 3B params, 128k context, closed API, $0.03/$0.15 per 1M tokens. ⚡ API live · OpenRouter. → Source
Schematron V2 Small — 3B-parameter HTML-to-JSON extractor prioritizing extraction quality for complex schemas and long pages. Released Sep 12 · 3B params, 128k context, closed API, $0.05/$0.23 per 1M tokens. ⚡ API live · OpenRouter. → Source
📊 Surfaced 3 · scanned 5 today
Saturday, September 12, 2026
Today's releases split between open-weight enterprise agentic models and cheap specialized extraction APIs, giving builders both control and low-cost structured data pipelines.
━━━ OPEN FRONTIER 🔓
Smaug — Open-weight models optimized for enterprise agentic AI, targeting builders who need controllable agent workflows without closed APIs. Released Sep 11 · Open weights. 📦 Open weights. → Source
━━━ SPECIALIZED 🎯
Schematron V2 Turbo — 3B-parameter HTML-to-JSON extractor tuned for throughput on high-volume extraction workloads. Released Sep 12 · 3B params, 128k context, closed API, $0.03/$0.15 per 1M tokens. ⚡ API live · OpenRouter. → Source
Schematron V2 Small — 3B-parameter HTML-to-JSON extractor prioritizing extraction quality for complex schemas and long pages. Released Sep 12 · 3B params, 128k context, closed API, $0.05/$0.23 per 1M tokens. ⚡ API live · OpenRouter. → Source
📊 Surfaced 3 · scanned 5 today
🤖 ModelBytes Digest
Sunday, September 13, 2026
Today's open-weight releases span multimodal, text, and audio with permissive licenses, giving builders composable components without API lock-in.
━━━ OPEN FRONTIER 🔓
Intern S2 397B — Image-text-to-text model with 397B total parameters under Apache 2.0, a large multimodal open-weight option for builders needing vision-language capability. Released Sep 13 · 397B params · Apache 2.0. 📦 Open weights · HF. → Source
Edge0 35B A3B preview — Text-generation preview with 35B total parameters under Apache 2.0, useful for testing new open-weight architectures before full release. Released Sep 13 · 35B params · Apache 2.0. 📦 Open weights · HF. → Source
━━━ SPECIALIZED 🎯
YuE2 3B — Text-to-audio model with 3B parameters under CC BY-NC 4.0, aimed at builders needing compact open-weight audio generation. Released Sep 13 · 3B params · CC BY-NC 4.0. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 5 today
Sunday, September 13, 2026
Today's open-weight releases span multimodal, text, and audio with permissive licenses, giving builders composable components without API lock-in.
━━━ OPEN FRONTIER 🔓
Intern S2 397B — Image-text-to-text model with 397B total parameters under Apache 2.0, a large multimodal open-weight option for builders needing vision-language capability. Released Sep 13 · 397B params · Apache 2.0. 📦 Open weights · HF. → Source
Edge0 35B A3B preview — Text-generation preview with 35B total parameters under Apache 2.0, useful for testing new open-weight architectures before full release. Released Sep 13 · 35B params · Apache 2.0. 📦 Open weights · HF. → Source
━━━ SPECIALIZED 🎯
YuE2 3B — Text-to-audio model with 3B parameters under CC BY-NC 4.0, aimed at builders needing compact open-weight audio generation. Released Sep 13 · 3B params · CC BY-NC 4.0. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 5 today
🤖 ModelBytes Digest
Monday, September 14, 2026
Today's pattern is infrastructure over raw scale: a production voice API, a 753B open-weight preview, and an attention-sparsification model each give builders a different lever for latency, cost, or control.
━━━ OPEN FRONTIER 🔓
Atria Dawn Preview — A 753B-parameter open-weight preview under MIT, sized for frontier-scale experimentation without a closed API. Released Sep 11 · 753.3B total params · MIT license. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
GPT-Live-1 — Natural voice experiences in the API, targeting real-time speech interactions for production voice agents. Released Sep 10. ⚡ API live · OpenAI. → Source
━━━ SPECIALIZED 🎯
Simple Attention Sparsification — Text-generation model from Tencent's attention-sparsification work, released as an open-weight artifact. Released Sep 14 · text-generation. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 12 today
Monday, September 14, 2026
Today's pattern is infrastructure over raw scale: a production voice API, a 753B open-weight preview, and an attention-sparsification model each give builders a different lever for latency, cost, or control.
━━━ OPEN FRONTIER 🔓
Atria Dawn Preview — A 753B-parameter open-weight preview under MIT, sized for frontier-scale experimentation without a closed API. Released Sep 11 · 753.3B total params · MIT license. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
GPT-Live-1 — Natural voice experiences in the API, targeting real-time speech interactions for production voice agents. Released Sep 10. ⚡ API live · OpenAI. → Source
━━━ SPECIALIZED 🎯
Simple Attention Sparsification — Text-generation model from Tencent's attention-sparsification work, released as an open-weight artifact. Released Sep 14 · text-generation. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 12 today
🤖 ModelBytes Digest
Tuesday, September 15, 2026
Today's releases split between Google's agent-cost push and Moonshot's 2.8T API debut, with a small Apache-2.0 local model filling the self-host gap.
━━━ CLOSED FRONTIER 🔒
Kimi K3 — Moonshot's 2.8T-parameter model hits Ropewalk, leading a 24-model weekly batch. 2.8T params. ⚡ API live · Ropewalk. → Source
Gemini 3.8 — Google's latest frontier release, framed by AI Business as a market-shift signal. → Source
Gemini 3.7 Flash — Positioned for lower-cost agent building, per Google's August update. → Source
━━━ LOCAL 🏠
NeoHorse 1 4B — A 4B Apache-2.0 text-generation model for local experimentation. Released Sep 15 · 4B params · Apache-2.0. 📦 Open weights · HF. → Source
📊 Surfaced 4 · scanned 2 today
Tuesday, September 15, 2026
Today's releases split between Google's agent-cost push and Moonshot's 2.8T API debut, with a small Apache-2.0 local model filling the self-host gap.
━━━ CLOSED FRONTIER 🔒
Kimi K3 — Moonshot's 2.8T-parameter model hits Ropewalk, leading a 24-model weekly batch. 2.8T params. ⚡ API live · Ropewalk. → Source
Gemini 3.8 — Google's latest frontier release, framed by AI Business as a market-shift signal. → Source
Gemini 3.7 Flash — Positioned for lower-cost agent building, per Google's August update. → Source
━━━ LOCAL 🏠
NeoHorse 1 4B — A 4B Apache-2.0 text-generation model for local experimentation. Released Sep 15 · 4B params · Apache-2.0. 📦 Open weights · HF. → Source
📊 Surfaced 4 · scanned 2 today
🤖 ModelBytes Digest
Wednesday, September 16, 2026
Today’s releases push builders toward real-time multimodal interaction: voice models that reason aloud, a free multimodal agentic API, and a robotics pose model.
━━━ CLOSED FRONTIER 🔒
Union Alpha — Multimodal model built for research, coding, and agentic workflows with frontier-level general performance. 262k context, closed/API, free. ⚡ API live · OpenRouter. → Source
Gemini 3.8 Live — Real-time speech-to-speech with reasoning that speaks as it works; Extended Thinking variant tops speech quality. Released Sep 16. ⚡ API live. → Source
━━━ SPECIALIZED 🎯
FoundationPose — Robotics pose estimation model from NVIDIA. Released Sep 15. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 4 today
Wednesday, September 16, 2026
Today’s releases push builders toward real-time multimodal interaction: voice models that reason aloud, a free multimodal agentic API, and a robotics pose model.
━━━ CLOSED FRONTIER 🔒
Union Alpha — Multimodal model built for research, coding, and agentic workflows with frontier-level general performance. 262k context, closed/API, free. ⚡ API live · OpenRouter. → Source
Gemini 3.8 Live — Real-time speech-to-speech with reasoning that speaks as it works; Extended Thinking variant tops speech quality. Released Sep 16. ⚡ API live. → Source
━━━ SPECIALIZED 🎯
FoundationPose — Robotics pose estimation model from NVIDIA. Released Sep 15. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 4 today
🤖 ModelBytes Digest
Thursday, September 17, 2026
Today's open-weight drop pairs a 27B multimodal generalist with two compact Apache-2.0 document models — builders get fresh local options for vision and document workflows without API lock-in.
━━━ OPEN FRONTIER 🔓
Swift Qwen3.8 27B — 27B image-text-to-text model released today, open weights for multimodal builders. 27B params · Released Sep 17. 📦 Open weights · HF. → Source
━━━ SPECIALIZED 🎯
WeVisDoc 4B — Apache-2.0 document vision model at 4B parameters, aimed at document understanding tasks. 4B params · Apache-2.0 · Released Sep 16. 📦 Open weights · HF. → Source
WeVisDoc 2B — Smaller 2B sibling for lighter document vision workloads, same Apache-2.0 license. 2B params · Apache-2.0 · Released Sep 16. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 5 today
Thursday, September 17, 2026
Today's open-weight drop pairs a 27B multimodal generalist with two compact Apache-2.0 document models — builders get fresh local options for vision and document workflows without API lock-in.
━━━ OPEN FRONTIER 🔓
Swift Qwen3.8 27B — 27B image-text-to-text model released today, open weights for multimodal builders. 27B params · Released Sep 17. 📦 Open weights · HF. → Source
━━━ SPECIALIZED 🎯
WeVisDoc 4B — Apache-2.0 document vision model at 4B parameters, aimed at document understanding tasks. 4B params · Apache-2.0 · Released Sep 16. 📦 Open weights · HF. → Source
WeVisDoc 2B — Smaller 2B sibling for lighter document vision workloads, same Apache-2.0 license. 2B params · Apache-2.0 · Released Sep 16. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 5 today
🤖 ModelBytes Digest
Friday, September 18, 2026
Today's releases cluster around efficiency and modality coverage: a 5.9 GB ternary Qwen derivative, an any-to-any realtime open model, and a multimodal composite API all target builders who need frontier behavior without the usual footprint.
━━━ OPEN FRONTIER 🔓
Realtime Venus — Any-to-any realtime model under Apache 2.0 for low-latency multimodal pipelines. Released Sep 16 · Apache 2.0. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Pareto — Multimodal composite for research, coding, and agentic workflows with 262K context. 262K context · Closed/API · $2.50/$7.50 per 1M. Released Sep 17. ⚡ API live · OpenRouter. → Source
━━━ SPECIALIZED 🎯
Qwen3.8 Omni Flash — Alibaba's flash omni variant for low-latency multimodal tasks. Released Sep 18. → Source
━━━ LOCAL 🏠
Ternary Bonsai 2 27B — Ternary-weight compression of Qwen3.8 27B at 1.76 effective bits per weight, keeping 98.2% of base aggregate score in 5.9 GB. 27B params · 262K context · text+image input · Apache 2.0. Released Sep 17. 📦 Open weights. → Source
📊 Surfaced 4 · scanned 8 today
Friday, September 18, 2026
Today's releases cluster around efficiency and modality coverage: a 5.9 GB ternary Qwen derivative, an any-to-any realtime open model, and a multimodal composite API all target builders who need frontier behavior without the usual footprint.
━━━ OPEN FRONTIER 🔓
Realtime Venus — Any-to-any realtime model under Apache 2.0 for low-latency multimodal pipelines. Released Sep 16 · Apache 2.0. 📦 Open weights · HF. → Source
━━━ CLOSED FRONTIER 🔒
Pareto — Multimodal composite for research, coding, and agentic workflows with 262K context. 262K context · Closed/API · $2.50/$7.50 per 1M. Released Sep 17. ⚡ API live · OpenRouter. → Source
━━━ SPECIALIZED 🎯
Qwen3.8 Omni Flash — Alibaba's flash omni variant for low-latency multimodal tasks. Released Sep 18. → Source
━━━ LOCAL 🏠
Ternary Bonsai 2 27B — Ternary-weight compression of Qwen3.8 27B at 1.76 effective bits per weight, keeping 98.2% of base aggregate score in 5.9 GB. 27B params · 262K context · text+image input · Apache 2.0. Released Sep 17. 📦 Open weights. → Source
📊 Surfaced 4 · scanned 8 today
🤖 ModelBytes Digest
Saturday, September 19, 2026
Today's concrete releases split between cheap closed multimodal speed and open-weight mid-size models, while frontier labs posture rather than ship.
━━━ OPEN FRONTIER 🔓
Xing4.0 29B A4B — 29B-parameter text-generation model under Apache-2.0, a mid-size open-weight option for builders needing a permissive base. Released Sep 19. 29B total params, Apache-2.0. 📦 Open weights · HF. → Source
Muse Glimmer 30B — Meta's 30B open-weight model from the September briefing, adding another mid-size option to the open ecosystem. 30B params, open weights. 📦 Open weights. → Source
━━━ CLOSED FRONTIER 🔒
GLM 5.3 FlashX — High-speed variant of GLM-5.3-Flash, native multimodal, up to 200 tokens/s. Released Sep 18. 1.0M context, Closed/API, $0.37/$1.25 per 1M. ⚡ API live · OpenRouter. → Source
━━━ SPECIALIZED 🎯
OrcaVerify Text 1.0 — Text verification model priced at $2.00/$0.00 per 1M tokens. Released Sep 16. $2.00/$0.00 per 1M tokens. ⚡ API live. → Source
📊 Surfaced 4 · scanned 8 today
Saturday, September 19, 2026
Today's concrete releases split between cheap closed multimodal speed and open-weight mid-size models, while frontier labs posture rather than ship.
━━━ OPEN FRONTIER 🔓
Xing4.0 29B A4B — 29B-parameter text-generation model under Apache-2.0, a mid-size open-weight option for builders needing a permissive base. Released Sep 19. 29B total params, Apache-2.0. 📦 Open weights · HF. → Source
Muse Glimmer 30B — Meta's 30B open-weight model from the September briefing, adding another mid-size option to the open ecosystem. 30B params, open weights. 📦 Open weights. → Source
━━━ CLOSED FRONTIER 🔒
GLM 5.3 FlashX — High-speed variant of GLM-5.3-Flash, native multimodal, up to 200 tokens/s. Released Sep 18. 1.0M context, Closed/API, $0.37/$1.25 per 1M. ⚡ API live · OpenRouter. → Source
━━━ SPECIALIZED 🎯
OrcaVerify Text 1.0 — Text verification model priced at $2.00/$0.00 per 1M tokens. Released Sep 16. $2.00/$0.00 per 1M tokens. ⚡ API live. → Source
📊 Surfaced 4 · scanned 8 today
🤖 ModelBytes Digest
Sunday, September 20, 2026
━━━ SPECIALIZED 🎯
Qwen Image 2.1 PE-I2I — A 9.4B-parameter image-to-image model extending Qwen’s open image stack for editing workflows. Released Sep 20 · 9.4B params. 📦 Open weights · HF. → Source
📊 Surfaced 1 · scanned 3 today
Sunday, September 20, 2026
━━━ SPECIALIZED 🎯
Qwen Image 2.1 PE-I2I — A 9.4B-parameter image-to-image model extending Qwen’s open image stack for editing workflows. Released Sep 20 · 9.4B params. 📦 Open weights · HF. → Source
📊 Surfaced 1 · scanned 3 today
🤖 ModelBytes Digest
Monday, September 21, 2026
Open-weight image and world models land today, giving builders local generation options across stills and spatial video.
━━━ SPECIALIZED 🎯
Qwen Image 2.1 — 7.1B text-to-image base now open, separate from the PE-I2I edit variant; a direct still-image generator for local pipelines. Released Sep 21 · 7.1B total params. 📦 Open weights · HF. → Source
WorldCrafter Base — 14.3B-param world model for generating consistent video from text or image prompts; the larger option for quality-first pipelines. Released Sep 21 · 14.3B total params. 📦 Open weights · HF. → Source
WorldCrafter Fast — Speed-tuned variant of the same world-model family, aimed at real-time or high-throughput generation where latency beats maximum fidelity. Released Sep 21. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 6 today
Monday, September 21, 2026
Open-weight image and world models land today, giving builders local generation options across stills and spatial video.
━━━ SPECIALIZED 🎯
Qwen Image 2.1 — 7.1B text-to-image base now open, separate from the PE-I2I edit variant; a direct still-image generator for local pipelines. Released Sep 21 · 7.1B total params. 📦 Open weights · HF. → Source
WorldCrafter Base — 14.3B-param world model for generating consistent video from text or image prompts; the larger option for quality-first pipelines. Released Sep 21 · 14.3B total params. 📦 Open weights · HF. → Source
WorldCrafter Fast — Speed-tuned variant of the same world-model family, aimed at real-time or high-throughput generation where latency beats maximum fidelity. Released Sep 21. 📦 Open weights · HF. → Source
📊 Surfaced 3 · scanned 6 today