Model comparison · 2026-09-22
Apertus v1.5 70B: when the 262K context and image input genuinely matter
Covers that Apertus v1.5 70B carries a context window four times wider than 70B Instruct plus image input at the same price, and which task types those two extras genuinely make a difference on.
Same price, two extra capabilities
publicai/apertus-v1.5-70b carries the exact same input/output price as publicai/apertus-70b-instruct ($0.82 / $2.92 per million tokens), but comes with two concrete additions: a 262,144-token context window (four times 70B Instruct's 65,536) and image input support alongside text (as a base64 data URL). Since the price is the same, there's no cost difference between the two if you're not using these two extra capabilities.
The catalog description positions v1.5 70B as producing stronger results than the 8B version on quality-sensitive work like information synthesis, analysis, multilingual assistant flows, content drafting, and coding help; that confirms it's the largest model in the family.
Which tasks the 262K context makes a difference on
A context window four times wider is decisive on tasks where 65K falls short: when you need to evaluate a large codebase, a long contract, or a research document drawing on many sources in a single request, v1.5 70B can process it as one piece, while 70B Instruct requires you to chunk the input. If your typical inputs stay under 65K tokens, this difference isn't felt in practice.
- Large codebase, long contract, multi-source document: the 262K context is decisive.
- Typical input under 65K: the context difference isn't felt in practice.
- If image input is needed, only v1.5 70B supports it.
If image input is needed, the decision is already made
If your flow needs to read visual content (a screenshot, a scanned document, a product photo) and evaluate it alongside text, 70B Instruct doesn't support that; in that case, v1.5 70B coming at the same price makes the decision easy. Images are accepted only as base64 data URLs — sending a remote image URL isn't supported.
Frequently asked questions
Is v1.5 70B always a better choice than 70B Instruct?
Since the price is the same, there's no downside to choosing v1.5 70B even if you don't use its extra capabilities (wide context, image input); the real question is whether you need those extras — see the 70B Instruct article.
Can I send the image as a remote URL?
No, the model accepts only a base64-encoded data URL format; sending a remote image URL isn't supported.