inference/research
verifiedhigh · 0.85updated 2026-10-16

Scaleway Generative APIs (serverless, Paris)

other

Fields

id
services/svc-scaleway-generative-apis
Provider
other
Endpoint
https://api.scaleway.ai/v1
Input $/Mtok
1.0128
Output $/Mtok
1.0128
Price basis
EUR per 1M TOKENS converted to USD, representative model llama-3.3-70b-instruct at €0.90 in / €0.90 out on the Paris serverless rate card, fetched 2026-10-16. USD fields use the ECB reference rate that day via Frankfurter: 1 EUR = 1.1253 USD, so €0.90 = $1.0128 (rounded to 4 d.p.). The EUR figure is the auditable one; the USD figure moves with the rate. Same card, same day, still in EUR per million: gpt-oss-120b €0.15/€0.60, qwen3-235b-a22b-instruct-2507 €0.75/€2.25, qwen3.6-35b-a3b €0.25/€1.50, deepseek-v4-flash-0731 €0.40/€0.80 (cached input €0.08), glm-5.3 €1.80/€5.50. Batch API is -50%. Prices are before tax. Free tier: first 1,000,000 tokens.
Context window
100,000
TTFT (ms)
1,330
Output tok/s
83
Measured by
artificial-analysis
Regions
fr-par
notes
The EU-residency cheap-API outlier that the Oct-14 seed missed: Scaleway serves the model from Paris, which is the point of the record, not the price. On the 2026-10-16 Artificial Analysis Llama 3.3 70B comparison it is the most expensive of 11 providers by blended price ($1.05 per 1M, up to 8.6x the DeepInfra floor) and mid-pack on speed: 83 output tok/s median, 1.33s median first chunk, 100k context — against Groq at 305.7 tok/s and DeepInfra at $0.10/$0.32. AA's $1.05 blended figure is consistent with €0.90/$1.0128 once input and output are blended; it is not a second price. provider='other' because the schema enum has no scaleway value; the record name carries it. Dedicated deployment is a different product and is GPU-hour priced (H100-1-80G €3.40/hr on the same page), which belongs in supply, not here. models_served lists only slugs present in data/models/.

Notes

The EU-residency cheap-API outlier that the Oct-14 seed missed: Scaleway serves the model from Paris, which is the point of the record, not the price. On the 2026-10-16 Artificial Analysis Llama 3.3 70B comparison it is the most expensive of 11 providers by blended price ($1.05 per 1M, up to 8.6x the DeepInfra floor) and mid-pack on speed: 83 output tok/s median, 1.33s median first chunk, 100k context — against Groq at 305.7 tok/s and DeepInfra at $0.10/$0.32. AA's $1.05 blended figure is consistent with €0.90/$1.0128 once input and output are blended; it is not a second price. provider='other' because the schema enum has no scaleway value; the record name carries it. Dedicated deployment is a different product and is GPU-hour priced (H100-1-80G €3.40/hr on the same page), which belongs in supply, not here. models_served lists only slugs present in data/models/.

Cross-references

Models (5)