fix(amazon-bedrock): correct Meta, Mistral, and Writer catalog

models/meta/llama-3.3-70b-instruct.toml

@@ -3,7 +3,7 @@ description = "Popular open Llama workhorse for multilingual chat, coding, and s
 family = "llama"
 release_date = "2024-12-06"
 last_updated = "2024-12-06"
-attachment = true
+attachment = false
 reasoning = false
 temperature = true
 tool_call = true

models/meta/llama-4-scout-17b-instruct.toml

@@ -1,3 +1,4 @@
+# https://github.com/meta-llama/llama-models/blob/main/models/llama4/MODEL_CARD.md
 name = "Llama 4 Scout 17B Instruct"
 description = "Open Llama with long-context vision for efficient multimodal agents"
 family = "llama"
@@ -11,7 +12,7 @@ knowledge = "2024-08"
 open_weights = true
 
 [limit]
-context = 3_500_000
+context = 10_000_000
 output = 16_384
 
 [modalities]

models/mistral/magistral-small-2509.toml

@@ -0,0 +1,27 @@
+# https://docs.mistral.ai/models/magistral-small-1-2-25-09
+# https://huggingface.co/mistralai/Magistral-Small-2509
+# https://huggingface.co/mistralai/Magistral-Small-2509/blob/main/generation_config.json
+# Native generation config sets max_new_tokens = 131072, matching the model card's max_tokens setting.
+name = "Magistral Small 1.2"
+description = "Open multimodal reasoning model for transparent analysis of text and images"
+family = "magistral"
+release_date = "2025-09-18"
+last_updated = "2025-09-18"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = true
+license = "Apache 2.0"
+
+[limit]
+context = 131_072
+output = 131_072
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/mistralai/Magistral-Small-2509"

models/mistral/ministral-3-14b-instruct-2512.toml

@@ -0,0 +1,29 @@
+# https://docs.mistral.ai/models/ministral-3-14b-25-12
+# https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512
+# https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/main/generation_config.json
+# Native examples request max_tokens = 262144; generation_config.max_length is also 262144.
+# This is a shared prompt/generation ceiling, not a host-specific independent output cap.
+name = "Ministral 3 14B"
+description = "Open vision-language model for efficient local deployment, instruction following, and tool use"
+family = "ministral"
+release_date = "2025-12-02"
+last_updated = "2025-12-02"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+license = "Apache 2.0"
+
+[limit]
+context = 262_144
+output = 262_144
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512"

models/mistral/ministral-3-3b-instruct-2512.toml

@@ -0,0 +1,29 @@
+# https://mistral.ai/news/mistral-3
+# https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512
+# https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/main/generation_config.json
+# Native examples request max_tokens = 262144; generation_config.max_length is also 262144.
+# This is a shared prompt/generation ceiling, not a host-specific independent output cap.
+name = "Ministral 3 3B"
+description = "Compact open vision-language model for edge deployment, instruction following, and tool use"
+family = "ministral"
+release_date = "2025-12-02"
+last_updated = "2025-12-02"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+license = "Apache 2.0"
+
+[limit]
+context = 262_144
+output = 262_144
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512"

models/mistral/ministral-3-8b-instruct-2512.toml

@@ -0,0 +1,29 @@
+# https://mistral.ai/news/mistral-3
+# https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512
+# https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/main/generation_config.json
+# Native examples request max_tokens = 262144; generation_config.max_length is also 262144.
+# This is a shared prompt/generation ceiling, not a host-specific independent output cap.
+name = "Ministral 3 8B"
+description = "Compact open vision-language model for edge deployment, instruction following, and tool use"
+family = "ministral"
+release_date = "2025-12-02"
+last_updated = "2025-12-02"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+license = "Apache 2.0"
+
+[limit]
+context = 262_144
+output = 262_144
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512"

models/mistral/mistral-large-2512.toml

@@ -1,7 +1,8 @@
+# https://mistral.ai/news/mistral-3
 name = "Mistral Large 3"
 description = "Mistral's largest general model for enterprise agents, coding, and multilingual reasoning"
 family = "mistral-large"
-release_date = "2024-11-01"
+release_date = "2025-12-02"
 last_updated = "2025-12-02"
 attachment = true
 reasoning = false

models/mistral/voxtral-mini-3b-2507.toml

@@ -0,0 +1,29 @@
+# https://mistral.ai/news/voxtral/
+# https://huggingface.co/mistralai/Voxtral-Mini-3B-2507
+# https://huggingface.co/mistralai/Voxtral-Mini-3B-2507/blob/main/params.json
+# https://docs.mistral.ai/api/endpoint/chat
+# Native params specify 32768 positions, matching the card's 32k context.
+# Output is this shared-window ceiling: prompt + max_tokens must fit; no separate native output cap is published.
+name = "Voxtral Mini 3B 2507"
+description = "Open audio-language model for speech transcription, audio understanding, and voice-driven tool use"
+family = "voxtral"
+release_date = "2025-07-15"
+last_updated = "2025-07-15"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+license = "Apache 2.0"
+
+[limit]
+context = 32_768
+output = 32_768
+
+[modalities]
+input = ["text", "audio"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507"

models/mistral/voxtral-small-24b-2507.toml

@@ -0,0 +1,29 @@
+# https://mistral.ai/news/voxtral/
+# https://huggingface.co/mistralai/Voxtral-Small-24B-2507
+# https://huggingface.co/mistralai/Voxtral-Small-24B-2507/blob/main/params.json
+# https://docs.mistral.ai/api/endpoint/chat
+# Native params specify 32768 positions, matching the card's 32k context.
+# Output is this shared-window ceiling: prompt + max_tokens must fit; no separate native output cap is published.
+name = "Voxtral Small 24B 2507"
+description = "Open audio-language model for speech transcription, audio understanding, and voice-driven tool use"
+family = "voxtral"
+release_date = "2025-07-15"
+last_updated = "2025-07-15"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+license = "Apache 2.0"
+
+[limit]
+context = 32_768
+output = 32_768
+
+[modalities]
+input = ["text", "audio"]
+output = ["text"]
+
+[[weights]]
+label = "Hugging Face"
+url = "https://huggingface.co/mistralai/Voxtral-Small-24B-2507"

models/writer/palmyra-x4.toml

@@ -1,7 +1,9 @@
+# https://writer.com/blog/actions-with-palmyra-x4/
+# https://dev.writer.com/home/models
 name = "Palmyra X4"
-description = "Reasoning model for deliberate analysis, multi-step problem solving, and tool use"
+description = "Enterprise language model for workflow automation, coding, data analysis, and tool use"
 family = "palmyra"
-release_date = "2025-04-28"
+release_date = "2024-10-09"
 last_updated = "2025-04-28"
 attachment = false
 reasoning = true
@@ -10,8 +12,8 @@ tool_call = true
 open_weights = false
 
 [limit]
-context = 122_880
-output = 8_192
+context = 128_000
+output = 4_096
 
 [modalities]
 input = ["text"]

models/writer/palmyra-x5.toml

@@ -1,18 +1,21 @@
+# https://writer.com/blog/long-context-palmyra-x5/
+# https://dev.writer.com/home/models
+# https://dev.writer.com/home/chat-with-images
 name = "Palmyra X5"
 description = "Reasoning model for deliberate analysis, multi-step problem solving, and tool use"
 family = "palmyra"
 release_date = "2025-04-28"
 last_updated = "2025-04-28"
-attachment = false
+attachment = true
 reasoning = true
 temperature = true
 tool_call = true
 open_weights = false
 
 [limit]
-context = 1_040_000
+context = 1_000_000
 output = 8_192
 
 [modalities]
-input = ["text"]
+input = ["text", "image"]
 output = ["text"]

providers/amazon-bedrock/models/eu.mistral.pixtral-large-2502-v1:0.toml

@@ -1,3 +1,4 @@
+# https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-pixtral-large.html
 # EU cross-Region inference profile, verified via ListInferenceProfiles from eu-west-1.
 # EU geo pricing (Price List eu-west-1 standard): input/output USD per 1M tokens.
 base_model = "mistral/pixtral-large-2502"

providers/amazon-bedrock/models/meta.llama3-1-70b-instruct-v1:0.toml

@@ -1,23 +1,8 @@
+# https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-1-70b-instruct.html
+# https://aws.amazon.com/bedrock/pricing/ — us-east-1 standard, USD per 1M tokens.
+base_model = "meta/llama-3.1-70b-instruct"
 name = "Llama 3.1 70B Instruct"
-description = "Open Llama instruction model for multilingual chat, reasoning, and coding"
-family = "llama"
-release_date = "2024-07-23"
-last_updated = "2024-07-23"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-12"
-tool_call = true
-open_weights = true
 
 [cost]
 input = 0.72
 output = 0.72
-
-[limit]
-context = 128_000
-output = 4_096
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/amazon-bedrock/models/meta.llama3-1-8b-instruct-v1:0.toml

@@ -1,23 +1,8 @@
+# https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-1-8b-instruct.html
+# https://aws.amazon.com/bedrock/pricing/ — us-east-1 standard, USD per 1M tokens.
+base_model = "meta/llama-3.1-8b-instruct"
 name = "Llama 3.1 8B Instruct"
-description = "Open Llama instruction model for multilingual chat, reasoning, and coding"
-family = "llama"
-release_date = "2024-07-23"
-last_updated = "2024-07-23"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-12"
-tool_call = true
-open_weights = true
 
 [cost]
 input = 0.22
 output = 0.22
-
-[limit]
-context = 128_000
-output = 4_096
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/amazon-bedrock/models/meta.llama3-3-70b-instruct-v1:0.toml

@@ -1,23 +1,8 @@
+# https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-3-3-70b-instruct.html
+# https://aws.amazon.com/bedrock/pricing/ — us-east-1 standard, USD per 1M tokens.
+base_model = "meta/llama-3.3-70b-instruct"
 name = "Llama 3.3 70B Instruct"
-description = "Open Llama instruction model for multilingual chat, reasoning, and coding"
-family = "llama"
-release_date = "2024-12-06"
-last_updated = "2024-12-06"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-12"
-tool_call = true
-open_weights = true
 
 [cost]
 input = 0.72
 output = 0.72
-
-[limit]
-context = 128_000
-output = 4_096
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/amazon-bedrock/models/meta.llama4-maverick-17b-instruct-v1:0.toml

@@ -1,23 +1,11 @@
-name = "Llama 4 Maverick 17B Instruct"
-description = "Open multimodal Llama model for strong reasoning and fast responses"
-family = "llama"
-release_date = "2025-04-05"
-last_updated = "2025-04-05"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-08"
-tool_call = true
-open_weights = true
+# https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-meta-llama-4-maverick-17b-instruct.html
+# Converse US profile verified 2026-09-09: maxTokens=8192 succeeds; 16384 exceeds the explicit 8192 limit.
+# https://aws.amazon.com/bedrock/pricing/ — us-east-1 standard, USD per 1M tokens.
+base_model = "meta/llama-4-maverick-17b-instruct"
 
 [cost]
 input = 0.24
 output = 0.97
 
 [limit]
-context = 1_000_000
-output = 16_384
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
+output = 8_192