OCT 27, 2026 · MAPLEHILL LABS · WEEKLY · MARKET
Model market week of October 26, 2026
The market this week
New in the index (5):
- GPT-6.1 Terra Pro (OpenAI) — from $1.00/MTok input
- GPT-6.1 Terra (OpenAI) — from $1.00/MTok input
- LFM2.5-2.6B (Liquid) — from $0.350/MTok input
- Kimi K3.1 Pro (Moonshot AI) — from $2.00/MTok input
- MiniMax M3.1 Flash (MiniMax) — from $0.142/MTok input
Price moves — largest 8 of 29 observed moves:
- Mistral Large 4 input: $0.680 → $1.36/MTok (+100.0%)
- Mistral Large 4 output: $2.09 → $4.18/MTok (+100.0%)
- Qwen3.8 27B input: $0.020 → $0.035/MTok (+75.0%)
- DeepSeek V4 Pro 0423 output: $1.10 → $0.416/MTok (-62.2%)
- DeepSeek V4 Pro 0423 input: $0.549 → $0.208/MTok (-62.2%)
- DeepSeek V4.1 Flash 1017 output: $0.080 → $0.120/MTok (+50.0%)
- DeepSeek V4 Flash 0423 input: $0.041 → $0.022/MTok (-45.0%)
- DeepSeek V4 Flash 0423 output: $0.086 → $0.047/MTok (-44.8%)
Provider availability:
- MiniMax M3.1 Flash: now on Relace, ModelRun, SambaNova, CheaperInference, DigitalOcean, GMICloud, Nebius, Novita, Wafer; left Relace, Wafer, Wafer
- Beam 1.0 501B A23B: now on CheaperInference, Io Net, DigitalOcean; left CheaperInference, Io Net
- Nemotron 3.5 Super VL: now on Nebius, Darkbloom, CheaperInference, Io Net, BaseTen
- Command A+: now on CheaperInference; left CheaperInference
- Hy4 preview: now on AtlasCloud, CheaperInference, Cohere
- Seed-2.0-Code: now on CheaperInference; left CheaperInference
- DeepSeek V4 Pro 0813: now on Nebius, BaseTen
- Aion-3.0-Mini: now on CheaperInference, CheaperInference; left CheaperInference, CheaperInference
Most-used models (OpenRouter tokens, trailing 7 days):
- DeepSeek V4.1 Flash 1017 — 35.4T tokens
- DeepSeek V4.1 Flash — 17.4T tokens
- GLM 5.3 Flash — 13.8T tokens
- Hy4 preview — 9.0T tokens
- GPT-6 Luna — 9.0T tokens
Measured note: z-ai/glm-5.2 on together slowed week-over-week — median TTFT 0.50s → 1.11s (≥3 daily probes each week).
Around the industry
- Google released TorchTPU, an open-source native PyTorch stack for TPUs, on October 20, 2026. [1]
- DeepSeek open-sourced its DeepInfer inference framework under the Apache 2.0 license on October 20, 2026. [1]
- Google Cloud, Red Hat, and IBM released llm-d 1.0, an open-source production-scale LLM inference stack, on October 20, 2026. [2]
- OpenAI and Google Cloud announced on October 21, 2026, that GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna, and Codex workflows would enter Google Cloud Model Garden private preview. [4]
- Anthropic launched Claude Duet, a research-preview agent that delivers a daily morning brief, on October 21, 2026. [1]
- MiniMax announced M3.1 Flash on October 19, 2026, claiming nearly three-times-faster inference and 70% lower agent-task costs than its predecessor. [6]
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.