The four current flagships, side by side

These are the four models ModelDex currently marks as the flagship of their family: GPT-5.5 from OpenAI, Claude Opus 4.8 from Anthropic, Gemini 3.5 Flash from Google, and Grok 4.3 from xAI. They are the models most teams will reach for when the work is serious. This guide compares them strictly on the verified specs we hold, so you can match one to your task without guessing.

A note up front: a flagship is the leading model of its line, not necessarily the priciest in absolute terms. As you will see, the four sit at very different price points, which is exactly why the choice between them is interesting.

Price

This is the sharpest dividing line. Grok 4.3 is the cheapest of the four by a wide margin at $1.25 per million input tokens and $2.50 per million output. Gemini 3.5 Flash is next at $1.50 input and $9.00 output. Claude Opus 4.8 is $5 input and $25 output. GPT-5.5 is $5 input and $30 output.

Put plainly, on output, the side that drives most bills, Grok 4.3 at $2.50 is one tenth the cost of GPT-5.5 at $30 and Gemini 3.5 Flash at $9.00 sits roughly in the middle. If your workload is high volume and the cheaper model clears your quality bar, the cost difference between these four is not a rounding error. It is a factor of ten or more.

Context window

All four carry very large windows, so for most tasks context is not the deciding factor. Gemini 3.5 Flash lists 1,048,576 tokens. GPT-5.5 lists 1,050,000. Claude Opus 4.8 and Grok 4.3 each list one million. These are close enough that you should choose on other axes unless you have a specific need that sits right at the boundary.

Maximum output

There is a real difference in how much each can write in a single response. Claude Opus 4.8 and GPT-5.5 each cap a single response at 128,000 output tokens. Gemini 3.5 Flash caps a single response at 65,536 tokens. If your task needs to generate a very long single artifact in one call, the higher output ceiling on Opus 4.8 and GPT-5.5 is worth noting. Grok 4.3 does not have a maximum output figure recorded in our dataset, so check the xAI documentation for that specific limit.

Reasoning

All four are reasoning models, but they expose the control differently. GPT-5.5 offers an explicit effort dial with levels of none, low, medium, high, and xhigh, which gives you the most granular control of the four. Claude Opus 4.8 uses adaptive thinking, adjusting its depth to the difficulty of the request. Gemini 3.5 Flash supports thinking mode. Grok 4.3 supports reasoning. If you want fine grained, per call control over how hard the model thinks, GPT-5.5's explicit five step dial is the most direct lever.

Inputs

All four accept text and image input. Gemini 3.5 Flash goes further on input modality, accepting text, image, video, audio, and PDF, which makes it the most flexible of the four when your inputs are not just text and pictures. If you are feeding the model video, audio, or PDFs directly, that breadth is a real differentiator. All four are text output models in our dataset.

How to choose between them

Choose Grok 4.3 when cost at volume is the priority and its quality clears your bar. At $1.25 input and $2.50 output it is dramatically cheaper than the other three and still a flagship-class model with a one million token window.

Choose Gemini 3.5 Flash when you need broad multimodal input, text plus image plus video plus audio plus PDF, at a mid price point of $1.50 input and $9.00 output. It is the most versatile of the four on what you can put into it.

Choose Claude Opus 4.8 when you want deep, adaptive reasoning and a large single response budget of 128,000 output tokens at $5 input and $25 output.

Choose GPT-5.5 when you want the most explicit control over reasoning effort, the same 128,000 output ceiling, and you are willing to pay the highest output rate of the four at $30 per million tokens.

For many teams the right answer is not one of these but a combination: a cheaper model as the default and a more expensive flagship reserved for the calls that need it. The point of comparing them on specs is to make that routing decision on evidence rather than reputation.

Where the numbers come from

Every price, window, output limit, reasoning capability, and modality in this comparison is a verified figure on the corresponding ModelDex model page, traced to the provider's own documentation: OpenAI, Anthropic, Google, and xAI. These models are recent and their specs can change as providers update their docs, so the live model pages are always the current truth.