Model market week of September 14, 2026

The market this week

New in the index (7):

• Schematron V2 Turbo (Inference-net) — from $0.030/MTok input
• Schematron V2 Small (Inference-net) — from $0.050/MTok input
• Fugu Ultra v2 (Sakana AI) — from $5.00/MTok input
• Fugu Max (Sakana AI) — from $2.00/MTok input
• Ling 3.0 Flash VL (InclusionAI) — from $0.060/MTok input
• DeepSeek V4.1 Flash (DeepSeek) — from $0.150/MTok input
• Mercury 2.5 (Inception) — from $0.040/MTok input

Price moves — largest 8 of 20 observed moves:

• Solar Pro 4 input: $0.030 → $0.090/MTok (+200.0%)
• Solar Pro 4 output: $0.120 → $0.360/MTok (+200.0%)
• DeepSeek V4 Flash 0731 output: $0.160 → $0.070/MTok (-56.3%)
• Hy3 output: $0.290 → $0.435/MTok (+50.0%)
• MiniMax M3 output: $0.960 → $0.480/MTok (-50.0%)
• Hy3 input: $0.070 → $0.105/MTok (+50.0%)
• MiniMax M3 input: $0.230 → $0.120/MTok (-47.8%)
• DeepSeek V4 Flash 0731 input: $0.050 → $0.035/MTok (-29.6%)

Provider availability:

• DeepSeek V4.1 Flash: now on Phala, Reka, Alibaba, BaseTen, Relace, Together, Modal, Parasail, SiliconFlow, Venice, Wafer, Fireworks, GMICloud, Io Net, Morph; left Io Net, Venice
• GLM 5.3 Flash: now on AtlasCloud, Io Net, Reka, Crusoe, Phala; left Io Net, Reka
• GLM 5.3: now on Io Net, Baidu; left Io Net, StreamLake
• Qwen3.8 27B: now on DeepInfra, DekaLLM, Mancer 2
• DeepSeek V4 Flash 0731: left Parasail, DeepSeek
• GLM 5.2: left Crusoe
• MiniMax M3: now on Mara
• MiMo-V2.5: now on Venice

Most-used models (OpenRouter tokens, trailing 7 days):

• GPT-5.6 Luna — 18.2T tokens
• Hy4 preview — 15.2T tokens
• DeepSeek V4 Flash 0731 — 11.7T tokens
• GLM 5.3 Flash — 11.6T tokens
• MiMo-V2.5 — 8.2T tokens

Measured note: deepseek/deepseek-v4-pro on deepinfra slowed week-over-week — median TTFT 0.81s → 1.19s (≥3 daily probes each week).

Around the industry

• DeepSeek released DeepSeek‑V4.1‑Flash, a natively multimodal Mixture‑of‑Experts model with a 552B backbone, 1M‑token context, and heavily compressed KV cache on September 10, 2026. [1]
• DeepSeek made DeepSeek‑V4.1‑Flash generally available as open weights under an MIT license on Hugging Face on September 10, 2026. [1]
• OpenAI released GPT‑6 Astra, described as its most capable model for work, and made it available in ChatGPT Work, Codex, and the API on September 10, 2026. [1] [8]
• IBM and NASA announced the open‑source release of the NASA‑IBM Lunar Foundation Model, a foundation model for scientific exploration of the Moon, on September 10, 2026. [10]
• IBM released Granite Time Series PatchTST‑FM‑r2, the latest model in its Granite TSFM time‑series family, on September 10, 2026. [1]
• OpenAI launched GPT‑Live‑1 in its API on September 10, 2026, providing a full‑duplex voice model that enables apps to listen and talk simultaneously, priced at $0.05 per minute. [9]
• Cohere Labs open‑sourced North Small Translate, a 218B Mixture‑of‑Experts translation model covering 50 languages and reported to beat DeepL on WMT26 benchmarks, on September 10, 2026. [9]
• IBM and NASA made the NASA‑IBM Lunar Foundation Model one of the first publicly available foundation models specifically targeting lunar scientific exploration, as part of their September 10, 2026 open‑source release. [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.