No Recorded Price Moves, But Host Gaps Remain Wide
Prices were unchanged in the weekly record, but the cheapest offers show sharp differences between hosts for the same models.
No price changes recorded
The clearest result this week is also the least dramatic: the price-moves list is empty. That means WhatModel recorded no model or host price increases or reductions during the last 7 days. I wouldn't turn that into a claim that pricing stood still everywhere; it means there were no recorded moves in this week's data.
The catalog did change. It now covers 224 models from 125 providers, with 12 new models added and 8 removed. The new additions range from free models to expensive batch offerings. LiquidAI's LFM2.5-2.6B and NVIDIA's Nemotron 3.5 Lightning are both listed at 0 for input and output. Upstage's Solar Pro 4 is listed at 0.03 input and 0.12 output.
At the other end, OpenAI's o1-pro (batch) is listed at 75 input and 300 output. OpenAI's o3 Pro (batch) comes in at 10 and 40, while GPT-5 Pro (batch) is listed at 7.5 input and 60 output. Anthropic's Claude Sonnet 4.6 (batch) is listed at 1.5 and 7.5.
Those are new entries, not price cuts. The distinction matters. A newly listed offer can change what buyers see without showing up as a price move.
The same model, very different hosts
The widest visible host differences are in the cheapest-offers list, especially for Mistral Nemo. DeepInfra lists it at 0.019 input and 0.03 output. Novita lists the same model at 0.04 and 0.17, while io-net lists it at 0.0429 and 0.165.
That is a meaningful spread in both directions. DeepInfra is the cheapest listed host for Mistral Nemo on input and output. Novita's output price is substantially higher than DeepInfra's, and io-net's input price is also higher. If a workload uses a lot of generated text, the output column deserves at least as much attention as the input price.
The list also shows why host comparisons can't stop at model names. OpenAI's GPT-5 Nano is listed at 0.025 input and 0.2 output on OpenAI, but Azure lists it at 0.05 and 0.4. The model label is the same; the listed host price is not.
Batch pricing adds another wrinkle. OpenAI's GPT-5 Nano (batch) has the same listed price as the regular GPT-5 Nano on OpenAI: 0.025 input and 0.2 output. Google Gemini 2.5 Flash Lite (batch) is listed at Google at 0.05 and 0.2, while Google AI Studio lists the non-batch Gemini 2.5 Flash Lite at 0.05 and 0.2.
Cheap models with quality attached
The best-value table favors low input prices, as expected. OpenAI's GPT-5.6 Luna is listed at 0.1 input with a quality score of 83.7 and a score of 87. DeepSeek's DeepSeek V4 Flash 0423 is listed at 0.14 input, with quality of 81.2 and a score of 86.
Those two entries lead the table's quality-per-dollar measure at 837 and 580. Nex AGI's Nex-N2-Pro follows at 0.25 input, with quality of 75.2 and a score of 79. The table also includes OpenAI's GPT-5.6 Terra at 1 input, quality of 90.5, and score of 90. Its value figure is 90.5, lower than the cheaper leaders but paired with the highest quality shown in this group.
The new models have no quality or score values yet. That makes it too early to judge whether the free additions or the new batch entries offer good value. For now, what I'd actually do is compare the host price first, then check quality once those records are populated. This week's biggest savings opportunity isn't a recorded cut. It's choosing the cheaper host before sending traffic.