Model market week of September 28, 2026

The market this week

New in the index (21):

• Claude Sonnet 5.5 (Anthropic) — from $2.00/MTok input
• Nex-N2.5-Mini (NEX AGI) — from $0.025/MTok input
• Nex-N2.5-Pro (NEX AGI) — from $0.075/MTok input
• Perceptron Mk1.5 (Perceptron) — from $0.150/MTok input
• Ember-1 (Fireworks) — from $3.00/MTok input
• Devstral 2 2512 (Mistral) — from $0.400/MTok input
• Mistral Large 3 2512 (Mistral) — from $0.500/MTok input
• GLM 5.3 Prime (Z.ai) — from $2.80/MTok input
• Qwen3.8 Max Prime (Qwen) — from $4.00/MTok input
• Space Bunny Alpha (Stealth) — from Free/MTok input
• Space Bunny Alpha (Stealth) — from Free/MTok input
• Aion 3.5 Mini (AION Labs) — from $0.700/MTok input
• Aion 3.5 (AION Labs) — from $3.00/MTok input
• Solar Mini 4 (Upstage) — from $0.050/MTok input
• Command A+ (Cohere) — from $0.300/MTok input
• GPT-6 Luna Pro (OpenAI) — from $0.100/MTok input
• GPT-6 Luna (OpenAI) — from $0.100/MTok input
• GPT-6 Sol Pro (OpenAI) — from $2.00/MTok input
• GPT-6 Sol (OpenAI) — from $2.00/MTok input
• Claude Opus 5.5 (Anthropic) — from $4.00/MTok input
• Qwen3.8 Omni Flash (Qwen) — from $0.150/MTok input

Price moves — largest 8 of 26 observed moves:

• DeepSeek V4 Pro 0423 output: $0.693 → $1.91/MTok (+175.7%)
• DeepSeek V4 Pro 0423 input: $0.346 → $0.955/MTok (+175.7%)
• DeepSeek V4 Flash 0423 input: $0.030 → $0.012/MTok (-60.0%)
• DeepSeek V4.1 Flash input: $0.100 → $0.050/MTok (-50.0%)
• DeepSeek V4.1 Flash output: $0.420 → $0.250/MTok (-40.5%)
• Nemotron 3.5 Lightning input: $0.065 → $0.039/MTok (-40.0%)
• Kimi K2.6 output: $1.72 → $2.39/MTok (+39.0%)
• Hy3 output: $0.528 → $0.330/MTok (-37.5%)

Provider availability:

• MiMo-V2.6-Flash: now on Novita, Relace, Venice, DeepInfra
• DeepSeek V4.1 Flash: now on Baidu, Ionstream, Krea, InferenceNet, NextBit, CheaperInference, DekaLLM, Krea, OpenInference; left Krea
• Hy4 preview: now on DeepInfra, Novita, SiliconFlow
• Ling 3.0 Flash Fin: now on Novita
• GLM 5.3: now on Relace, Makora, CheaperInference, PrimeIntellect; left Io Net, Makora
• DeepSeek V4 Pro 0813: now on Io Net, Relace, CheaperInference; left Fireworks, Sail Research, BaseTen
• Muse Glimmer 30B: left Fireworks
• DeepSeek V4 Flash 0731: now on CheaperInference, Cohere; left Fireworks

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

• DeepSeek V4.1 Flash — 20.8T tokens
• Space Bunny Alpha — 18.2T tokens
• GLM 5.3 Flash — 13.3T tokens
• Hy4 preview — 9.0T tokens
• GPT-5.6 Luna — 8.5T tokens

Measured note: x-ai/grok-4.5 on xai slowed week-over-week — median TTFT 0.76s → 2.14s (≥3 daily probes each week).

Around the industry

• Anthropic released Claude Opus 5.5 on September 22, 2026, as part of its frontier Claude 5.5 lineup. [4] [8]
• Anthropic released Claude Sonnet 5.5 on September 28, 2026, extending the Claude 5.5 family to a mid‑tier model. [4]
• OpenAI’s GPT-6 Sol and GPT-6 Luna models were released on September 22, 2026, following the earlier launch of GPT‑6 Astra. [8] [9]
• OpenAI cut API prices for updated GPT-6 Sol and GPT-6 Luna on September 23, 2026, to $2/$10 and $0.10/$0.50 per million input/output tokens respectively, about half or less than prior GPT‑5.6 pricing, with the company stating the reductions are permanent. [11]
• Google’s Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS text-to-speech models became generally available via the Gemini API on September 22, 2026, with the launch noted in a September 23 blog update. [8]
• Alibaba’s Qwen team released Qwen-Audio-3.1 on September 23, 2026, adding upgraded ASR, TTS, and realtime models plus new ASR-Next and TTS-Next variants. [7]
• Alibaba released Qwen3.8-Omni-Flash on September 18, 2026, as part of its September model refresh that also included Qwen-Audio-3.1 later in the month. [8]
• xAI launched Grok 4.7 on September 21, 2026, adding another frontier-scale model to the September 2026 release wave. [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.