Mistral Small 3.2 is live — pin mistral-small-2506 before -latest bites you in prod

ApiDelta · 2026-05-15 · 298 mots · apidelta.maxiaworld.app

🚨 Breaking

Nothing today. No breaking changes across tracked providers.

🗑️ Deprecations

No new deprecation deadlines. Worth noting: mistral-small-latest will eventually resolve to mistral-small-2506 — no forced migration deadline announced yet, but alias-pinned production deployments will absorb the swap silently.

💰 Pricing

No pricing changes announced alongside the Small 3.2 rollout. Mistral Small historically sits below GPT-4o-mini and competes directly with DeepSeek V2.5 in the mid-tier. Before routing prod traffic to the new model ID, verify your rate limits at console.mistral.ai — new model IDs occasionally start with tighter quotas.

🆕 Nouveautés

Mistral Small 3.2 — mistral-small-2506 now callable via API

Mistral shipped the model on June 20 (changelog) and flipped the API switch on June 23 (changelog).

Key facts: - Model ID: mistral-small-2506 - Endpoint unchanged: https://api.mistral.ai/v1/chat/completions - Drop-in swap from mistral-small-2503

Migration is a one-liner if you pin IDs (you should):

# before
model = "mistral-small-2503"
# after — explicit, no alias dependency
model = "mistral-small-2506"

No regression reports in the wild yet. Run your eval suite on a shadow deployment before cutting over.

💡 Conseil du jour

Thin changelog day. Use it to audit *-latest aliases across every provider in your stack — OpenAI, Anthropic, and Mistral all use them. A silent model swap under an alias is one of the quietest ways to ship a regression that won't show up in your diff.

Action today: grep -r 'latest' ./config ./src | grep -v '.git' — any hit in a production code path gets replaced with an explicit dated model ID before you close your laptop.

#api#llm#en#mistral#model-release