SEP 1, 2026 · MAPLEHILL LABS · WEEKLY · MARKET
Model market week of August 31, 2026
The market this week
New in the index (4):
- Granite 4.2 8B (IBM Granite) — from $0.100/MTok input
- Hy4 preview (Tencent) — from $0.834/MTok input
- Qwen3.8 Flash (Qwen) — from $0.150/MTok input
- GLM 5.3 Flash (Z.ai) — from $0.071/MTok input
Price moves — largest 8 of 20 observed moves:
- DeepSeek V4 Flash 0731 output: $0.090 → $0.160/MTok (+78.0%)
- DeepSeek V4 Flash 0731 input: $0.030 → $0.050/MTok (+66.7%)
- GPT-5.6 Sol input: $4.00 → $2.00/MTok (-50.0%)
- GPT-5.6 Sol output: $20.00 → $10.00/MTok (-50.0%)
- GLM 5.2 output: $1.07 → $1.56/MTok (+45.3%)
- GLM 5.2 input: $0.342 → $0.487/MTok (+42.7%)
- GLM 5.3 Flash output: $0.167 → $0.237/MTok (+42.5%)
- GLM 5.3 Flash input: $0.050 → $0.071/MTok (+42.4%)
Provider availability:
- GLM 5.3 Flash: now on StreamLake, Fireworks, DigitalOcean, Friendli, Makora, Morph, Phala, Reka, Relace, SiliconFlow, Together, Wafer
- GLM 5.3: now on Decart, Venice, Wafer, Reka, SiliconFlow, Phala, AkashML, AtlasCloud, BaseTen, Cloudflare, DeepInfra, DigitalOcean, Fireworks, Friendli, GMICloud, Io Net, Modal, Morph, Novita, Parasail, Together; left AtlasCloud
- DeepSeek V4 Pro 0813: now on NextBit, CoreWeave, Sail Research; left Sail Research
- KAT-Coder-Pro V2.5: now on AtlasCloud; left StreamLake
- Kimi K2.7 Code: now on StreamLake, Nebius, Fireworks; left Nebius, Parasail, Fireworks
- GLM 5.1: left Parasail
- Trinity Large Thinking: left Parasail
- DeepSeek V3.2: now on Mara
Most-used models (OpenRouter tokens, trailing 7 days):
- DeepSeek V4 Flash 0731 — 12.2T tokens
- GPT-5.6 Luna — 8.5T tokens
- GLM 5.3 Flash — 8.1T tokens
- MiMo-V2.5 — 8.1T tokens
- Hy3 — 6.4T tokens
Measured note: z-ai/glm-5.2 on novita slowed week-over-week — median TTFT 1.53s → 5.78s (≥3 daily probes each week).
Around the industry
- OpenAI announced new partnerships bringing ChatGPT for Teachers to 55 additional U.S. school systems across 20 states, expanding free access and training to over 300,000 educators and staff and keeping the tool free for verified K–12 educators through June 2028. [9]
- OpenAI published an incident report on the Hugging Face compromise and updated its safety policies to require chain-of-thought monitoring for all tool-using reinforcement learning training and evaluations involving models with GPT‑5.6 Sol capability or higher. [5]
- Anthropic opened a research preview of its Model Hardware Standard (MHS), a shared specification for AI agents to safely operate physical devices, to an initial group of scientific research labs and advanced manufacturers. [6]
- Anthropic’s Claude Developer Platform released Compliance API session endpoints out of beta for Cowork and Claude Code sessions and expanded local session transcripts support to Claude Science and Microsoft 365 sessions for Claude Enterprise. [3]
- Anthropic updated the Claude Developer Platform to add personal keys and service account keys in the Claude Console for workspace-scoped access and tracking, and aligned SDK behavior for files and skills including renaming BetaSkill to BetaContainerSkill. [10]
- Anthropic released Claude Code version 2.1.247, adding a SendFeedback tool to draft feedback reports, a /claude-api cost-optimize command to help reduce project spending, and multiple safeguards and usability improvements across plugins, sessions, and remote control features. [2] [3]
- Google DeepMind launched Gemini 3.5 Transcribe, a real-time, context-aware speech-to-text model for live streaming and pre-recorded audio pipelines, alongside Gemini Omni 1.1 Flash for more fine-grained multimodal video and scene generation controls in Google AI Studio and for developers. [4] [8]
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.