SEP 8, 2026 · MAPLEHILL LABS · WEEKLY · MARKET
Model market week of September 7, 2026
The market this week
New in the index (10):
- GPT-6 Astra (OpenAI) — from $10.00/MTok input
- GPT-6 Astra Pro (OpenAI) — from $10.00/MTok input
- Qwen3.8 Max (0902) (Qwen) — from $2.00/MTok input
- Ling 3.0 Flash Fin (InclusionAI) — from $0.060/MTok input
- Nemotron 3.5 Content Safety (NVIDIA) — from $0.200/MTok input
- Muse Spark 1.3 Contributor (Meta) — from $0.100/MTok input
- Muse Spark 1.3 (Meta) — from $1.25/MTok input
- Gemini 3.8 Flash (Google) — from $0.750/MTok input
- Claude Fable 5.1 (Anthropic) — from $10.00/MTok input
- Mercury 2.5 Preview (Inception) — from $0.040/MTok input
Price moves — largest 8 of 17 observed moves:
- Qwen3.8 27B output: $0.200 → $2.00/MTok (+900.0%)
- DeepSeek V4 Flash 0731 output: $0.100 → $0.160/MTok (+60.0%)
- Qwen3.8 27B input: $0.220 → $0.150/MTok (-31.8%)
- Step 3.7 Flash input: $0.200 → $0.160/MTok (-20.0%)
- Step 3.7 Flash output: $1.15 → $0.920/MTok (-20.0%)
- DeepSeek V4 Pro 0423 input: $0.870 → $0.748/MTok (-14.0%)
- MiniMax M2.7 input: $0.240 → $0.210/MTok (-12.5%)
- MiniMax M2.7 output: $0.960 → $0.840/MTok (-12.5%)
Provider availability:
- GLM 5.3 Flash: now on Relace, CoreWeave, Sail Research, NextBit; left Phala, Relace
- DeepSeek V4 Pro 0813: now on Ionstream, Baidu
- Nex-N2-Pro: left SiliconFlow
- DeepSeek V4 Pro 0423: left Ionstream, CoreWeave
- Qwen3.5-9B: now on Darkbloom
- Kimi K2.5: left DeepInfra
- GLM 5.3: now on Crusoe, StreamLake, AtlasCloud, Inceptron, Makora, Sail Research
- GLM 5.2: left Reka, Sail Research
Most-used models (OpenRouter tokens, trailing 7 days):
- Hy4 preview — 17.1T tokens
- GPT-5.6 Luna — 14.2T tokens
- DeepSeek V4 Flash 0731 — 12.3T tokens
- GLM 5.3 Flash — 12.2T tokens
- DeepSeek V4 Flash 0423 — 5.1T tokens
Measured note: z-ai/glm-5.2 on z-ai slowed week-over-week — median TTFT 1.32s → 3.88s (≥3 daily probes each week).
Around the industry
- OpenAI launched the GPT-6 Astra model to business users in its Daybreak testing program on September 3, 2026. [14]
- Anthropic released Claude Fable 5.1 and restricted-access Claude Mythos 5.1 on September 1, 2026. [2]
- Google released Gemini 3.8 Flash and Gemini 3.8 Flash Cyber on September 2, 2026. [2]
- Meta released Muse Spark 1.3 and Muse Spark 1.3 Contributor on September 2, 2026. [2]
- Alibaba released the Qwen3.8-Max-0902 model on September 2, 2026. [2]
- The AI release tracker LLM Gateway lists GPT-6 Astra as the most recent of six new AI models released so far in September 2026, with Astra’s release dated September 3, 2026. [12]
- An AI model catalog describes GPT-6 Astra as OpenAI’s flagship computer-use and coding model, released September 3, 2026, rather than a chat-focused model. [10]
- The same catalog states Claude Mythos 5.1 shares its underlying model with Claude Fable 5.1 but is released with safeguards lifted for verified cybersecurity and life-science work, including reduced refusal rates in cyber tasks and advanced protein-design capabilities. [10]
Assembled automatically from ModelIndex's observed data and cited external sources. Sourced claims link to their citations; our own figures link to the live pages they come from.