H3 Image vs Seedream vs GPT Image vs Nano Banana: What You Can Call Today (2026)

People will search H3 Image versus Seedream 5.5 versus GPT Image versus Nano Banana. Two of those names still have no SandBase model ID.

H3 Image versus Seedream versus GPT Image versus Nano Banana is the comparison the last two launch weeks invite, and it is the wrong comparison to ship on. On 10 November 2026 MiniMax Code still calls H3 Image upcoming. ByteDance’s 9 November note prefers Seedream 5.5 to GPT Image 2.5 Sunburst and Nano Banana 2.1, and Seedream 5.5 has no API ID either. Sunburst and Nano Banana 2.1 do have cards. So does Seedream 5.0. A tester’s game-asset thread is not a model string.

If the job is “a still today,” start on Seedream 5.0 Pro. If the job is the Sunburst or Nano Banana 2.1 named in ByteDance’s chart, open those cards and run your prompt. Do not wait on an H3 Image route that has not been published, and do not treat Hailuo Early Access anecdotes as the test.

Key takeaway

• H3 Image has no Open Platform ID and no SandBase card on 10 November 2026. Hailuo’s web-app config lists MiniMax-H3-Image with a 16-reference cap and filter tags edit, 4k, and art. Tester talk of 1K/2K/4K and 21:9–9:16 stays reported.
• Seedream 5.5 also returned HTTP 404 on SandBase. The callable Seedream image cards are 5.0 Pro at $0.09 per image and 5.0 Flash at $0.018 per image on the billing table. The catalog header still says $0.02 for Flash.
• GPT Image 2.5 Sunburst’s list table marks high/1k at $0.09, with a 25% off badge. JSON-LD on that page offers $0.0675. The run control displayed $0.09. Read the button at order time.
• Nano Banana 2.1 shows a 15% off badge. JSON-LD offers $0.034 per run. The payload list rate at the default 1K is $0.04, and the run control displayed $0.04.
• Neither ByteDance’s 9 November chart nor our October poster test measured H3 Image.

Where these names sit

The stills you can call are 5.0, Sunburst, and Nano Banana 2.1. H3 on SandBase is video. This series is linked at the end.

• What is confirmed about H3 Image holds the Code page, the Hailuo config, and the 404 list.
• Seedream 5.5 release and Seedream 5.5 versus Sunburst and Nano Banana 2.1 keep the vendor chart away from routes you can call. Seedream 5.0 Pro’s introduction is the live generation.
• GPT Image 2.6 shipped later than ByteDance’s chart. Nano Banana 2.1’s October status note recorded a gap the SandBase card has since closed. The September Flare versus Sunburst note compared documentation, not this week’s models. Images 2.5 versus Nano Banana 2 is an older pair.
• Cards: Seedream 5.0 Flash, GPT Image 2.5 Sunburst, Nano Banana 2.1. The MiniMax catalog is where H3 Image is absent and H3 video is present. Calls use the run API.

A routing table with two blank price cells is the useful artifact. A score would be fiction.

Who can actually be called

Checked 10 November 2026, about 08:39 UTC. “Price we will quote” means a figure printed on a page that day, with the page named. It is not a forecast.

SandBase MiniMax catalog on 10 Nov 2026, Asia/Shanghai. Two-region crop of the same page: the vendor header counts 0 image records and 15 video records (the page’s own snapshot label reads Oct 10, 2026), and the grid shows the H3 / H3 Max video family (text-to-video, image-to-video, reference-to-video). There is no still-image H3 card.

Sunburst’s card publishes this list-style table. Quality down the side, resolution across. High is marked as the default. Failed runs on that card are not charged.

The same page showed a 25% off badge. Structured data offers $0.0675 per run, which is 25% under the high/1k cell. The run control displayed $0.09. llms-full.txt says “From $0.07 per run (25% off list price).” I am not collapsing those into one charged price. Open the card when you order.

SandBase GPT Image 2.5 Sunburst card on 10 Nov 2026, Asia/Shanghai. Two-region crop of the same page: the 25% off badge and id openai/gpt-image-2.5-sunburst, plus a visible run estimate of $0.0675 and a RUN $0.09 button (25% off).

Nano Banana 2.1’s card describes a fast image model with cleaner text, extreme aspect ratios, and 1K to 4K output. That sentence is SandBase catalog copy retrieved that day. The page payload’s price formula, at default 1K with search off, is $0.04; 2K is $0.06; 4K is $0.135; web search or image search adds $0.014. Those are payload list rates. The visible badge says 15% off. JSON-LD offers $0.034, which is 15% under $0.04. The run control displayed $0.04. Catalog line: “From $0.03 per run (15% off list price).” Resolution values on the formula are 1K, 2K, and 4K, default 1K.

SandBase Nano Banana 2.1 card on 10 Nov 2026, Asia/Shanghai (google/nano-banana-2.1). Two-region crop of the same page: 15% off badge, copy that mentions 1K to 4K, a Resolution control listing 1K, 2K, 4K, a displayed run estimate of $0.034, and RUN $0.04.

Tighter crop of the Seedream 5.0 Pro card billing/run area on 10 Nov 2026, Asia/Shanghai: RUN $0.09 and “Displayed run estimate: $0.09.” This is 5.0 Pro, not Seedream 5.5 (still 404) and not H3 Image.

One thousand Seedream 5.0 Pro images at $0.09 is $90. One thousand Flash images at $0.018 is $18. There is no matching line for H3 Image or Seedream 5.5. I would not budget the year on the badge math alone.

Edit cards, same day, for jobs that start from a picture:

Hailuo’s 16-reference figure is the H3 Image app config. It is not the cap on these edit cards. SandBase H3 reference-to-video caps images at 9, and that route returns video.

What Hailuo and testers suggest H3 Image is for

This section is labeled so it cannot leak into the price table.

Hailuo’s own config, retrieved 10 November 2026, puts MiniMax-H3-Image first in imageModelKeys, type t2i, mode image-reference, maxSupportImageCount 16, prompt length 32,000, enableAt true, filter tags edit, 4k, art. The description string is “New architecture, stunning visuals, versatile styles.” MiniMax Code’s preview text points at CG and game concept art, instruction-guided editing, and spatial relationships, and then says the preview is not public access.

Testers, from the evening of 9 November 2026 Asia/Shanghai, posted game assets, 3D environments, concept art, VFX, oil paint, Eastern aesthetics, and CAD-style exploded views. One post cited 1K, 2K, and 4K, aspects from 21:9 to 9:16, and a 6272×2688 sample. One post said a 2K 16:9 frame cost 18 Hailuo credits. One called edit the strongest mode. @Hailuo_AI replied “thank you!!!!” and did not publish a spec. I would use that pile to decide which prompt to try if you already have Early Access. I would not use it to pick a SandBase route, because the route does not exist.

Dated boxes that did not measure H3 Image

ByteDance, on MagicArena, 9 November 2026, with their own evaluators: Seedream 5.5 leads GPT Image 2.5 Sunburst and Nano Banana 2.1 on text structure for both text-to-image and image-to-image, with a substantial text-to-image realism advantage. The only printed GSB percentage in that prose is +55.62% for image-to-image realism versus Seedream 5.0 Pro, not versus Sunburst or Nano Banana. Severe text corruption is given as 0.4%, minor 1.7%. The 5.5 comparison keeps the caveats: vendor-run, no prompt list, not GPT Image 2.6, and not a SandBase ID. H3 Image is not in that chart.

What SandBase published on 3 October 2026 is a different test. Five e-commerce poster prompts, two images each, exact-string grading. The writeup records Seedream 5.0 Pro, GPT Image 2.5 Flare, and Nano Banana 2 each at 10 of 10, with other models tied. Sunburst was not in it. Nano Banana 2.1 was not in it. Seedream 5.5 did not exist in public yet. H3 Image was not in it. Seedream 5.0 Pro’s list price on 10 November is $0.09, not the $0.0675 billed in October. Keep the old figure as a dated bill.

I would not publish a ranking that subtracts October pass rates from November GSB points and then inserts a November tester thread. The units do not match. The models do not match.

GPT Image 2.6 is the newer OpenAI tier. On 10 November the Flare card’s high/1k cell was $0.09, high/2k $0.15, high/4k $0.27, with a 5% off badge and a run control of $0.09. The Sunburst 2.6 card showed the same badge and a $0.09 run control. Do not merge 2.6 into ByteDance’s 5.5 chart. The 2.6 post is that pair. GPT Image 2 still lists high/1k at $0.22 with a 25% off badge. Different row.

How to choose

The trade-off is boring and useful. Pro at $0.09 is five times the Flash billing price, which tells you the invoice, not which picture is better. Sunburst’s high/4k list cell is $0.27, three times high/1k, before the badge. Nano Banana’s payload 4K list rate is $0.135, against $0.04 at 1K. H3 Image has no cell. Paying “about the same as Pro” because a tester liked a frame is how a blank cell becomes a fake budget.

Try today

API status, 10 November 2026: there is no H3 Image body to send. This sample calls Seedream 5.0 Pro. The quoted strings are the test, including a bilingual line. The Chinese characters live only inside the JSON.

curl -sS https://api.sandbase.ai/v1/run \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bytedance/seedream/5.0/pro",
    "prompt": "Editorial product still of a matte-black field monitor on a workbench, readable on-image typography, bilingual caption SAFE TO MOVE / 可以移动, soft sidelight, square crop.",
    "aspect_ratio": "1:1",
    "num_images": 1,
    "output_format": "png"
  }'

The same shape on Nano Banana 2.1, default resolution, search left off:

curl -sS https://api.sandbase.ai/v1/run \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/nano-banana-2.1",
    "prompt": "Editorial product still of a matte-black field monitor on a workbench, readable on-image typography, bilingual caption SAFE TO MOVE / 可以移动, soft sidelight, square crop.",
    "resolution": "1K",
    "aspect_ratio": "1:1",
    "output_format": "png"
  }'

Poll GET /v1/run/{id} until the status is terminal, as the run reference describes. Read the strings yourself. For Chinese, a person reads them. The poster workflow is that checklist on the live 5.0 cards. This article does not claim a fresh generation.

I would ship the still on Pro, Sunburst, or Nano Banana 2.1, and I would keep H3 Image in a folder named “early access, no ID” until a primary page prints a price.

FAQ

Which is cheaper, H3 Image or Seedream 5.0?

H3 Image has no published USD price. Seedream 5.0 Pro is $0.09 per image and Flash is $0.018 on the billing table, both on SandBase on 10 November 2026. image-01 at $0.0035 is MiniMax’s older image API, not H3 Image.

Can I call Seedream 5.5 or H3 Image on SandBase?

Not on 10 November 2026. Both families’ candidate paths returned HTTP 404. Call 5.0 Pro, 5.0 Flash, GPT Image 2.5 Sunburst, or Nano Banana 2.1.

Did anyone benchmark H3 Image against Nano Banana 2.1?

Not in a primary source checked for this note. ByteDance’s chart names Seedream 5.5, Sunburst, and Nano Banana 2.1. It does not name H3 Image. Tester posts are anecdotes with timestamps in the release note.

Should I use GPT Image 2.6 instead of 2.5 Sunburst?

2.6 is live and later than the chart. Use it when you have read its card. Do not paste 2.6 results backward into ByteDance’s 9 November GSB.

What if I need motion, not a still?

Leave this table. H3 text-to-video and H3 image-to-video are the live MiniMax routes. They are video.

Related reading

• H3 Image: what is confirmed: Code versus Hailuo config versus the missing card.
• Stills into H3 video shots: a temporary motion path.
• Seedream 5.5 routing guide: the vendor chart, still without a 5.5 ID.

The blank cells stay blank

Open Seedream 5.0 Pro, GPT Image 2.5 Sunburst, or Nano Banana 2.1 and run the prompt you would have sent to H3 Image. Archive the file with the model ID in the name. That set is the comparison that will matter once H3 Image can be named in a request.