diff --git a/packages/api/src/utils/tokens.spec.ts b/packages/api/src/utils/tokens.spec.ts index 62cfaed13f..1ba04b5915 100644 --- a/packages/api/src/utils/tokens.spec.ts +++ b/packages/api/src/utils/tokens.spec.ts @@ -75,6 +75,52 @@ describe('Bedrock NVIDIA Nemotron context windows', () => { }); }); +describe('Bedrock vendor-prefixed context windows', () => { + it.each([ + ['ai21.jamba-1-5-large-v1:0', 256000], + ['ai21.jamba-1-5-mini-v1:0', 256000], + ['anthropic.claude-mythos-preview', 1000000], + ['google.gemma-3-4b-it', 128000], + ['google.gemma-3-12b-it', 128000], + ['google.gemma-3-27b-it', 128000], + ['minimax.minimax-m2', 1000000], + ['minimax.minimax-m2.1', 196000], + ['minimax.minimax-m2.5', 196000], + ['mistral.devstral-2-123b', 256000], + ['mistral.magistral-small-2509', 128000], + ['mistral.ministral-3-3b-instruct', 128000], + ['mistral.ministral-3-8b-instruct', 128000], + ['mistral.ministral-3-14b-instruct', 128000], + ['mistral.voxtral-mini-3b-2507', 32768], + ['mistral.voxtral-small-24b-2507', 32768], + ['openai.gpt-5.4', 950000], + ['openai.gpt-5.5', 950000], + ['openai.gpt-6.1-terra', 950000], + ['openai.gpt-oss-safeguard-20b', 128000], + ['openai.gpt-oss-safeguard-120b', 128000], + ['qwen.qwen3-32b-v1:0', 32768], + ['qwen.qwen3-235b-a22b-2507-v1:0', 262144], + ['qwen.qwen3-coder-30b-a3b-v1:0', 262144], + ['qwen.qwen3-coder-480b-a35b-v1:0', 262144], + ['qwen.qwen3-coder-next', 262144], + ['qwen.qwen3-next-80b-a3b', 256000], + ['qwen.qwen3-vl-235b-a22b', 256000], + ['writer.palmyra-vision-7b', 4096], + ['writer.palmyra-x4-v1:0', 128000], + ['writer.palmyra-x5-v1:0', 128000], + ['xai.grok-4.6', 500000], + ['xai.grok-4.7', 524288], + ['zai.glm-4.7', 203000], + ['zai.glm-4.7-flash', 203000], + ['zai.glm-5', 200000], + ['zai.glm-5.3', 1000000], + ])('resolves %s to its Bedrock context window', (model, context) => { + for (const prefix of ['', 'us.', 'global.']) { + expect(getModelMaxTokens(`${prefix}${model}`, EModelEndpoint.bedrock)).toBe(context); + } + }); +}); + describe('getModelMaxTokens partial-override fallback', () => { const partialOverride: EndpointTokenConfig = { 'custom-model': { prompt: 1, completion: 2, context: 32000, output: 4096 }, diff --git a/packages/api/src/utils/tokens.ts b/packages/api/src/utils/tokens.ts index c7f54d1ceb..70ce36cb51 100644 --- a/packages/api/src/utils/tokens.ts +++ b/packages/api/src/utils/tokens.ts @@ -195,6 +195,7 @@ const anthropicModels = { 'claude-opus-5': 1000000, 'claude-fable-5': 1000000, 'claude-mythos-5': 1000000, + 'claude-mythos-preview': 1000000, 'claude-fable-5-1': 1000000, 'claude-mythos-5-1': 1000000, }; @@ -414,6 +415,48 @@ const openAIBedrockModels = { 'openai.gpt-6-astra': bedrockOpenAIContext, 'openai.gpt-6-sol': bedrockOpenAIContext, 'openai.gpt-6-luna': bedrockOpenAIContext, + 'openai.gpt-5.4': bedrockOpenAIContext, + 'openai.gpt-5.5': bedrockOpenAIContext, + 'openai.gpt-6.1-terra': bedrockOpenAIContext, + 'openai.gpt-oss-safeguard-20b': 128000, + 'openai.gpt-oss-safeguard-120b': 128000, +}; + +/* Bedrock serves these under vendor-prefixed ids that no generic key reaches, + so each needs its own entry to avoid the 32K fallback. + https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html */ +const vendorBedrockModels = { + 'ai21.jamba-1-5-large': 256000, + 'ai21.jamba-1-5-mini': 256000, + 'google.gemma-3-4b-it': 128000, + 'google.gemma-3-12b-it': 128000, + 'google.gemma-3-27b-it': 128000, + 'minimax.minimax-m2': 1000000, + 'minimax.minimax-m2.1': 196000, + 'minimax.minimax-m2.5': 196000, + 'mistral.devstral-2-123b': 256000, + 'mistral.magistral-small-2509': 128000, + 'mistral.ministral-3-3b-instruct': 128000, + 'mistral.ministral-3-8b-instruct': 128000, + 'mistral.ministral-3-14b-instruct': 128000, + 'mistral.voxtral-mini-3b-2507': 32768, + 'mistral.voxtral-small-24b-2507': 32768, + 'qwen.qwen3-32b': 32768, + 'qwen.qwen3-235b-a22b-2507': 262144, + 'qwen.qwen3-coder-30b-a3b': 262144, + 'qwen.qwen3-coder-480b-a35b': 262144, + 'qwen.qwen3-coder-next': 262144, + 'qwen.qwen3-next-80b-a3b': 256000, + 'qwen.qwen3-vl-235b-a22b': 256000, + 'writer.palmyra-vision-7b': 4096, + 'writer.palmyra-x4': 128000, + 'writer.palmyra-x5': 128000, + 'xai.grok-4.6': 500000, + 'xai.grok-4.7': 524288, + 'zai.glm-4.7': 203000, + 'zai.glm-4.7-flash': 203000, + 'zai.glm-5': 200000, + 'zai.glm-5.3': 1000000, }; // NVIDIA Nemotron models on Bedrock @@ -435,6 +478,7 @@ const bedrockModels = { ...amazonModels, ...openAIBedrockModels, ...nvidiaBedrockModels, + ...vendorBedrockModels, ...anthropicModels, }; diff --git a/packages/data-schemas/src/methods/tx.spec.ts b/packages/data-schemas/src/methods/tx.spec.ts index 09225a5ebd..ecdf506ecf 100644 --- a/packages/data-schemas/src/methods/tx.spec.ts +++ b/packages/data-schemas/src/methods/tx.spec.ts @@ -3272,6 +3272,35 @@ describe('Bedrock NVIDIA Nemotron pricing', () => { }); }); +describe('Bedrock vendor-prefixed pricing', () => { + it.each([ + ['ai21.jamba-1-5-large', 'ai21.jamba-1-5-large-v1:0', 2.0, 8.0], + ['ai21.jamba-1-5-mini', 'ai21.jamba-1-5-mini-v1:0', 0.2, 0.4], + ['mistral.magistral-small-2509', 'mistral.magistral-small-2509', 0.5, 1.5], + ['mistral.ministral-3-3b-instruct', 'mistral.ministral-3-3b-instruct', 0.1, 0.1], + ['mistral.ministral-3-8b-instruct', 'mistral.ministral-3-8b-instruct', 0.15, 0.15], + ['mistral.ministral-3-14b-instruct', 'mistral.ministral-3-14b-instruct', 0.2, 0.2], + ['openai.gpt-6.1-terra', 'openai.gpt-6.1-terra', 1.1, 5.5], + ['writer.palmyra-vision-7b', 'writer.palmyra-vision-7b', 0.15, 0.6], + ['writer.palmyra-x4', 'writer.palmyra-x4-v1:0', 2.5, 10.0], + ['writer.palmyra-x5', 'writer.palmyra-x5-v1:0', 0.6, 6.0], + ])('bills %s at its Bedrock rate, not the default', (key, model, prompt, completion) => { + for (const id of [model, `us.${model}`]) { + expect(getValueKey(id)).toBe(key); + expect(getMultiplier({ model: id, tokenType: 'prompt' })).toBe(prompt); + expect(getMultiplier({ model: id, tokenType: 'completion' })).toBe(completion); + } + }); + + it('bills the GPT-6.1 Terra global profile below the regional one', () => { + const model = 'global.openai.gpt-6.1-terra'; + expect(getValueKey(model)).toBe(model); + expect(getMultiplier({ model, tokenType: 'prompt' })).toBe(1.0); + expect(getMultiplier({ model, tokenType: 'completion' })).toBe(5.0); + expect(getMultiplier({ model: 'eu.openai.gpt-6.1-terra', tokenType: 'prompt' })).toBe(1.1); + }); +}); + describe('Grok long-context premium tier', () => { it('doubles the rate once a prompt passes the 200K threshold', () => { for (const model of ['grok-4.5', 'grok-4.6']) { diff --git a/packages/data-schemas/src/methods/tx.ts b/packages/data-schemas/src/methods/tx.ts index 3b27c3e8df..3bc6045d1b 100644 --- a/packages/data-schemas/src/methods/tx.ts +++ b/packages/data-schemas/src/methods/tx.ts @@ -114,6 +114,18 @@ const bedrockValues: Record = { 'nvidia.nemotron-nano-12b-v2': { prompt: 0.2, completion: 0.6 }, 'nvidia.nemotron-nano-3-30b': { prompt: 0.06, completion: 0.24 }, 'nvidia.nemotron-super-3-120b': { prompt: 0.15, completion: 0.65 }, + 'ai21.jamba-1-5-large': { prompt: 2.0, completion: 8.0 }, + 'ai21.jamba-1-5-mini': { prompt: 0.2, completion: 0.4 }, + 'mistral.magistral-small-2509': { prompt: 0.5, completion: 1.5 }, + 'mistral.ministral-3-3b-instruct': { prompt: 0.1, completion: 0.1 }, + 'mistral.ministral-3-8b-instruct': { prompt: 0.15, completion: 0.15 }, + 'mistral.ministral-3-14b-instruct': { prompt: 0.2, completion: 0.2 }, + /* Regional inference profiles (`us.`, `eu.`) bill 10% over the global profile. */ + 'openai.gpt-6.1-terra': { prompt: 1.1, completion: 5.5 }, + 'global.openai.gpt-6.1-terra': { prompt: 1.0, completion: 5.0 }, + 'writer.palmyra-vision-7b': { prompt: 0.15, completion: 0.6 }, + 'writer.palmyra-x4': { prompt: 2.5, completion: 10.0 }, + 'writer.palmyra-x5': { prompt: 0.6, completion: 6.0 }, }; /**