
Grok Imagine Video 1.5 Lite 対 Hailuo 2.3:「ライト」モデルの方が高い
- openaiNEWOpenAI: GPT-6.1 Sol2026-09-2952知能
- anthropicNEWAnthropic: Claude Sonnet 5.52026-09-2856知能
- typesafeTypeSafe: Jev 1.132026-09-24$0.04 / $0.00 100万トークンあたり · 111 tok/s
- OpenAIOpenAI: GPT-6 Luna2026-09-2238知能
- OpenAIOpenAI: GPT-6 Sol2026-09-2248知能
- AnthropicAnthropic: Claude Opus 5.52026-09-2258知能
- xAIGrok 4.72026-09-2146知能
- OrcaOrca: OrcaCyber Zero 1.02026-09-17$3.00 / $7.50 100万トークンあたり · 55 tok/s
- OrcaOrca: OrcaVerify Text 1.02026-09-16$2.00 / $0.00 100万トークンあたり · 347 tok/s
- DeepSeekDeepSeek: DeepSeek V4.1 Flash2026-09-1040知能
- OpenAIOpenAI: GPT-6 Astra2026-09-0453知能77コーディング
- GoogleGoogle: Gemini 3.8 Flash2026-09-0241知能76コーディング
- AlibabaQwen: Qwen3.8 Max (0902)2026-09-0245知能76コーディング
- AnthropicAnthropic: Claude Fable 5.12026-09-0153知能82コーディング
- TencentTencent: Hy4 preview2026-08-28$0.83 / $2.50 100万トークンあたり · 60 tok/s
- AlibabaQwen: Qwen3.8 Flash2026-08-26$0.15 / $0.47 100万トークンあたり · 377 tok/s
- z-aiZ.ai: GLM 5.3 Flash2026-08-2642知能72コーディング
- DeepSeekDeepSeek: DeepSeek V4 Flash Vision (Exp)2026-08-21$0.22 / $0.66 100万トークンあたり · 231 tok/s
- z-aiZ.ai: GLM 5.32026-08-1845知能75コーディング
- obsidianQwen3.8 27B2026-08-1534知能68コーディング
Start with the number that breaks the assumption. Grok Imagine Video 1.5 Lite is the budget tier of SpaceXAI's Grok Imagine Video family and it is quoted at $8.40 per minute on Artificial Analysis's AA-Video-T2V v2.0 board. Hailuo 2.3, MiniMax's video model released in April 2026, is quoted by resellers at roughly $0.08 per second — about $4.80 a minute, or a little over half of Lite's rate. In the matchup between the cheap Grok and Hailuo 2.3, the cheap Grok is the more expensive model. That is the whole comparison in one line, and everything else on this page is about why the sticker gap is smaller and stranger than it looks, and why the two models are not actually competing for the same clip.
二つの価格、大きく異なる二つの契約
The per-second comparison is real but it hides the shape of the bill. Grok Imagine Video 1.5 Lite is billed per second of output with no floor and no commitment — a $8.40-per-minute rate means a six-second draft costs $0.84 and a failed generation is not billed at all. Hailuo 2.3's official channel works the other way: MiniMax sells prepaid points packages running from $1,000 to $6,000, and those points burn at a fixed rate — one point for a six-second 768p standard clip, two points for a ten-second 768p or six-second 1080p clip, with the faster Hailuo-2.3-Fast variant dropping those to 0.7 and roughly 1.1 to 1.3. Resellers quote the flatter $0.067 to $0.08 per second instead, which is where the familiar "Hailuo is dirt cheap" number comes from.
So the honest version of the price row is this: if you are experimenting, the Grok model has the better commercial shape and Hailuo has the better rate; if you already buy MiniMax points at volume, Hailuo is the cheaper clip and it is not close. Neither arrangement is wrong. They are priced for different sizes of customer, and a solo creator comparing a $0.84 draft to a $1,000 minimum is not really comparing rates at all.
次元ごとに
• Price — Grok Imagine Video 1.5 Lite at $8.40 per minute, billed per second with no minimum, against Hailuo 2.3 at roughly $0.067–$0.08 per second through resellers or MiniMax's prepaid points packages starting at $1,000. Hailuo wins the rate; Lite wins the entry cost.
• Resolution — Lite renders 1080p at 720p and upscales it; Hailuo 2.3 produces native 1080p, and MiniMax has not attached an upscale caveat to it. On this row Hailuo is unambiguously ahead.
• Clip length — 1 to 15 seconds for Lite at any supported resolution, against a 6-second ceiling at 1080p and 10 seconds at 768p for Hailuo 2.3. Lite's window is more than twice as long at full resolution.
• Audio — the Grok Imagine Video family ships native audio on its full model, and Lite's own listing does not document an audio channel separately, so treat sound as unverified on Lite. Hailuo 2.3 has no native audio at all; an earlier Hailuo generation had it and 2.3 dropped it. Neither model gives you a soundtrack you can rely on without checking.
• Frame control — Lite accepts a first-frame image, and the family's full model documents first-, last- and mid-frame pinning, so a sequence can be chained by design. Hailuo 2.3 has no end-frame control, which is what makes multi-shot continuity on it a manual stitching job.
• Aspect ratios — seven on the Grok side (16:9, 9:16, 1:1, 4:3, 3:4, 3:2, 2:3). MiniMax does not present Hailuo 2.3 as an aspect-ratio-first model; its identity is its look, not its framing menu.
• Independent ranking — Grok Imagine Video 1.5 Lite sits at #17 on the v2.0 with-audio board at Elo 993 (±10, 5,715 samples). Hailuo 2.3 was ranked around #28 at roughly 1,180 Elo on an earlier version of the Artificial Analysis video board and does not appear in the v2.0 top 29, so the two are measured on different editions of the same board and their Elo numbers are not directly comparable. Anyone putting 993 and 1,180 side by side is comparing two different scales.
• Style — Lite inherits a generalist's range; Hailuo 2.3 is a stylist. Cinematic camera motion, human motion and micro-expressions, and stylized or anime content are its strengths, and MiniMax is upfront that photorealism is not the target.

Where Hailuo actually wins
Strip the pricing games out and Hailuo 2.3 has two genuine advantages over Grok Imagine Video 1.5 Lite, and both are about the pixel rather than the price.
The first is native resolution. Hailuo's 1080p is 1080p. Lite's 1080p is a 720p render carried up. For a clip that will sit in a feed that distinction evaporates; for a clip that will be cropped, keyed or graded it is the entire ballgame, and it is the reason Lite's lower per-minute rate is not the free lunch the tier name suggests.
The second is the look itself. Hailuo 2.3's camera motion and human micro-expression work is its reason for existing, and it produces a specific aesthetic — stylized, animated, deliberate — that a generalist model will not reproduce on request. If that is the look your project needs, no amount of ranking advantage on the Grok side substitutes for it, because you are not buying a score, you are buying a house style.

Liteがどこで勝ち、どれだけの差で勝つのか
クリップの長さは最初にして最大の要素です。解像度を問わず15秒と、1080pでの6秒は、単なる差ではありません。別種のショットです。15秒の窓はビート、リビール、短いシークエンスを支え、6秒の窓は一瞬を支え、それより長いものはつなぎ合わせになります。これにGrokファミリーのフレームピン留めを組み合わせれば、マルチショット・シークエンスは後から編集ソフトで組み立てるのではなく、APIに設計として組み込まれています。
2つ目は商用利用のハードルの低さです。最低利用額なし、秒単位の課金、失敗した生成には課金されず、始める前に購入量を見積もらなければならないポイントパッケージもありません。「このモデルは自分のコンテンツにそもそも使えるのか」を最初に問う人にとって、それは午後いっぱいのテストに終わるか、購入を決められるかの違いになります。
The third is breadth. Seven aspect ratios, text-to-video and image-to-video, first-frame input and reference images: Lite is a generalist's toolkit at a discount, and it is the cheaper way into a family whose full model was the strongest general-purpose video model in this series at launch. Hailuo is a specialist, and specialists are the right answer only when the specialty is what you need.

実際にどちらかを購入する方法
Neither model is served by OrcaRouter. We host neither Grok Imagine Video 1.5 Lite nor Hailuo 2.3, and both are reached through the vendor's own API and several third-party platforms — Lite per second through SpaceXAI's API, Hailuo 2.3 through MiniMax's points packages and resellers. This is the part of the stack where the routing question genuinely does not apply, and it is worth saying so rather than implying otherwise.
実際に効いてくるのは、クリップより上流にあるすべてです。ショットを指定するプロンプト、10 本のショットを順序づけるショットリスト、書き出しに添えて渡されるキャプションとメタデータ、40 本のドラフトのうちどれを採用するかを決めるパス——これがテキストモデルの仕事であり、OrcaRouter 上では プロバイダーの定価を 0% のマークアップでそのまま通す、200+ モデルにまたがる単一の API、プロバイダー間の自動フェイルオーバー、複数のモデルを 1 回の呼び出しにまとめるルーティング DSLとして動作します。どちらの動画ベンダーの契約もその層を変えることはありません。それこそが、より安いクリップを探すにはルーターが間違った場所であり、その周辺のより安いパイプラインを探すには正しい場所である理由です。
決定
Buy Hailuo 2.3 when you need native 1080p at the lowest per-second rate available and you want a specific stylized look with cinematic movement — and when a prepaid points package is a purchase you are prepared to make. Buy Grok Imagine Video 1.5 Lite when you need clips longer than six seconds, when you want to pay per second with no minimum while you find out whether the model suits your content, and when a 1080p string in your request is a label rather than a delivery specification.
The pairing only looks strange because the names mislead. "Lite" describes Lite's position inside its own family, not its position against the market, and Hailuo 2.3 has been undercutting most of the field on rate since April 2026. Read the two rows of this comparison as a specialist model with a cheap rate and a generalist model with a cheap entry, and the choice becomes obvious — it just is not the choice the tier names imply.
