fix: add Bedrock models mistral.devstral-2-123b and mistral.magistral-small-2509

[github-actions]

fix: add Bedrock models mistral.devstral-2-123b and mistral.magistral-small-2509

Closes #658

Source issue: #658

Summary

Field	Value
Provider	bedrock
Primary model	mistral.devstral-2-123b
Changed models	mistral.devstral-2-123b mistral.magistral-small-2509
Added models	mistral.devstral-2-123b mistral.magistral-small-2509
Updated models	None
Verification sources	1 2 3 [4](https://aws.amazon.com/bedrock/pricing/](https://aws.amazon.com/bedrock/pricing/)

Verified metadata

Model	Display name	Parent	Providers	Format	Flavor	Token limits	Pricing	Lifecycle
mistral.devstral-2-123b	Devstral 2 123B		bedrock	converse	chat	input=256000, output=32000	n/a	active
mistral.magistral-small-2509	Magistral Small (2509)		bedrock	converse	chat	input=128000, output=40000	n/a	multimodal=true; reasoning=true

Verification notes

Verification

mistral.devstral-2-123b

• Source 1: Bedrock model card — Confirms model ID mistral.devstral-2-123b, 256K context window, 32K max output tokens, text-only input/output, active lifecycle, Standard/Priority/Flex tiers. No reasoning support. Available in 14 regions (us-east-1, us-east-2, us-west-2, eu-central-1, eu-north-1, eu-south-1, eu-west-1, eu-west-2, ap-northeast-1, ap-south-1, ap-southeast-2, ap-southeast-3, sa-east-1, ap-southeast-4).
  • Verified fields: format (converse — Converse API supported), flavor (chat), max_input_tokens (256000), max_output_tokens (32000), displayName ("Devstral 2 123B" from model card title), multimodal (false — no image input), available_providers (bedrock)
• Source 2: Mistral AI models docs — Lists Devstral 2 as a production "frontier code agents model". Confirms it is an active, non-deprecated model.

mistral.magistral-small-2509

• Source 1: Bedrock model card — Confirms model ID mistral.magistral-small-2509, 128K context window, 40K max output tokens, reasoning supported, image input supported. Active lifecycle on Bedrock, Standard/Priority/Flex tiers. Available in 14 regions (us-east-1, us-east-2, us-west-2, eu-south-1, eu-west-1, eu-west-2, ap-northeast-1, ap-south-1, ap-southeast-2, sa-east-1, ap-southeast-3, ap-southeast-4, eu-central-1, eu-north-1).
  • Verified fields: format (converse — Converse API supported), flavor (chat), max_input_tokens (128000), max_output_tokens (40000), displayName ("Magistral Small (2509)" matching existing Mistral-platform entry style), multimodal (true — image input supported), reasoning (true — "Reasoning: Supported" on model card), available_providers (bedrock)
• Source 2: Mistral AI models docs — Lists magistral-small-2509 as deprecated on Mistral's own platform (retirement July 31, 2026). However, the Bedrock model card shows lifecycle as "Active" with EOL "N/A", so no deprecation metadata is set for the Bedrock entry.

Fields NOT verified

• input_cost_per_mil_tokens / output_cost_per_mil_tokens — Bedrock pricing page (https://aws.amazon.com/bedrock/pricing/) could not be fully scraped (page too large). Pricing fields are omitted entirely rather than guessed.
• input_cache_read_cost_per_mil_tokens / input_cache_write_cost_per_mil_tokens — Not applicable; Bedrock does not expose prompt caching pricing for these models.
• parent — No Bedrock-specific parent alias exists for either model (e.g., no mistral.devstral-2-latest or mistral.magistral-small-latest in model_list.json). The cross-platform Mistral entries are different provider/format and not valid parents.
• supported_regions — Not applicable; these are Bedrock models, not Vertex.
• locations — Not required for Bedrock models (only required for Vertex/publishers/ models per requiresExplicitLocations).
• deprecation_date — Not set. Both models have "Active" lifecycle and "N/A" EOL on Bedrock.

sync_models (LiteLLM) cross-check

Neither mistral.devstral-2-123b nor mistral.magistral-small-2509 (nor any Bedrock-prefixed variants) exist in the LiteLLM sync_models catalog. All proposed values come directly from official Bedrock model cards. No deviations to document since sync_models has no data for comparison.

sync_models vs proposed update

sync_models cross-check found differences. Official provider verification was used for the applied values, and sync_models discrepancies are listed below for review.

Model	Field	Proposed update	sync_models	sync_models source models
mistral.devstral-2-123b	max_output_tokens	32000	8192	mistral.devstral-2-123b
mistral.devstral-2-123b	input_cost_per_mil_tokens	n/a	0.4	mistral.devstral-2-123b
mistral.devstral-2-123b	output_cost_per_mil_tokens	n/a	2	mistral.devstral-2-123b
mistral.magistral-small-2509	max_output_tokens	40000	8192	mistral.magistral-small-2509
mistral.magistral-small-2509	input_cost_per_mil_tokens	n/a	0.5	mistral.magistral-small-2509
mistral.magistral-small-2509	output_cost_per_mil_tokens	n/a	1.5	mistral.magistral-small-2509

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
ai-proxy	Ready	Preview, Comment	May 24, 2026 10:41am