OpenAI floods the market with batch models

OpenAI supplied all 12 new listings in WhatModel’s market over the last 7 days. Every new entry is a batch version, from o1-pro and o3 to GPT-5 Codex and the GPT-5.6 family.

That makes this a release story, but not quite a normal one. The new listings have no quality scores yet, so buyers can compare prices and context windows, but not measured performance. The market now contains 224 models from 125 providers, and there were no recorded price moves.

OpenAI adds range, not much evidence yet

The new OpenAI entries cover a wide price range. o4 Mini (batch) and o3 Mini (batch) are listed at 0.55 input and 2.2 output, while o3 (batch) is listed at 1 input and 4 output. o1-pro (batch) rises to 75 input and 300 output.

The GPT-5.6 listings are more varied. GPT-5.6 Luna (batch) is listed at 0.1 input and 0.6 output, with a 1,050,000-token context window. Terra (batch) is listed at 1 input and 6 output. Sol and Sol Pro (batch) are each listed at 2.5 input and 15 output. The Pro versions don't always mean a higher listed input price: Terra Pro (batch) is also listed at 1 input and 6 output, while Sol Pro matches Sol.

The other notable additions are GPT-5 Codex (batch), with a 400,000-token context window and prices of 0.625 input and 5 output, plus GPT-5 Pro (batch), listed at 7.5 input and 60 output. The o-series entries have 200,000-token context windows. None has a score yet, so I wouldn't treat the labels alone as a buying recommendation.

Eight models leave the tracker

The removals are more mixed. Google’s Gemini 3.1 Pro Preview (batch), DeepSeek V4 Flash Latest, MoonshotAI’s Kimi K2.7 Code (batch), and Z.ai’s GLM 5.2 (batch) are gone from the current listings.

The list also includes Thinking Machines’ Inkling Small and Inkling (batch), NVIDIA’s Nemotron 3 Ultra (batch), and inclusionAI’s Ling 3.0 Tiny (free). Removal doesn't tell us why a model disappeared. It may reflect availability, a catalog change, or a provider decision. For anyone choosing a model, though, it is a reminder to check whether a supposedly cheap option is still offered by a host before building around it.

The value picks are mostly elsewhere

The strongest value result is GPT-5.6 Luna, with a quality rating of 85.3, a score of 88, and quality per dollar of 853. Its listed input price is 0.1. That ranking is striking, but it needs a caveat: the new listing is GPT-5.6 Luna (batch), and the value table names GPT-5.6 Luna without the batch label. Those may not be interchangeable offers.

DeepSeek V4 Flash 0423 follows with quality of 80.8, a score of 86, and quality per dollar of 577.1 at 0.14 input. GPT-5.6 Terra also stands out: quality is 91.1, score is 90, and quality per dollar is 91.1 at 1 input. Nano Banana 2 has quality of 80.6 and a score of 79 at 0.5 input, though its quality per dollar is 161.2.

For the lowest listed input prices, Mistral Nemo on deepinfra leads at 0.019 input and 0.03 output. Nex-N2-Mini on nex-agi is listed at 0.025 input and 0.1 output, while GPT-5 Nano on openai has the same input price and 0.2 output. Qwen3.7 Flash on alibaba is listed at 0.03 input and 0.13 output.

What I'd actually do this week is separate batch economics from live serving and wait for scores on the new OpenAI entries. For a measured value pick, GPT-5.6 Terra and DeepSeek V4 Flash 0423 have evidence behind them. For a low-cost workload, the cheapest offers deserve a closer look, but availability matters as much as the sticker price.