Request Budget

Before switching APIs

Keep the model ID, endpoint, features and serving settings in the same decision.

A familiar request shape has limits

OpenAI

Tool calling
Requires Responses API for GPT-6.1 Sol

gpt-6.1-sol

Context 1,050,000 and max output 128,000 tokens. Do not infer Chat Completions tool parity from an OpenAI-compatible request shape.

Source ↗

Cerebras

Completion choices
Only n=1 supported

Cerebras Chat Completions

max_completion_tokens includes reasoning. Reasoning effort defaults and support vary by model; compatible message format does not imply full API parity.

Source ↗

Anthropic

Fable 5.2 migration
Changing an earlier prefix can cause a 400 error

Fable/Mythos 5.2 thinking blocks reused after system/tools/earlier-message edits

Every account subject to prefix check. Earlier computer_20251124 tool also returns 400 except on Bedrock; migrate to computer_toolset_20260801. Model switch does not guarantee API feature parity.

Source ↗

Model identity is not a test result

DeepInfra

GPT-OSS-120B
$0.037 input / $0.17 output per 1M tokens

openai/gpt-oss-120b, public model listing, bfloat16

131,072 context listed. Zero retention badge alone is not a reviewed endpoint policy. No independent serving-fidelity or latency test.

Source ↗

Together AI

Hosted-model fidelity
Together says it hosts at full precision

Non-passthrough models hosted on Together

Vendor statement only. No independent output-fidelity test. Do not infer identical behavior across hosts or precision from the base model name.

Source ↗

Groq

Model access
Llama 3.3 70B is listed Enterprise / Contact Sales

llama-3.3-70b-versatile

Do not apply old public rates to this listing. Preview models are evaluation-only and may be discontinued at short notice.

Source ↗

A dated condition can change the plan

Google Vertex AI

Flash introductory tokens
$0.75 input / $3.75 output per 1M tokens

Gemini 3.8, 3.7 and 3.6 Flash; CodeMender using these models

USD introductory rates through December 31, 2026. Announced January 1, 2027 standard rates $1.50 input / $7.50 output. Separate model from the GPT-OSS illustration.

Source ↗

Google Vertex AI

Failed long completions
Gemini 4 retrievable usage can still be charged

Gemini 4 completion above 32K tokens

Effective November 13, 2026: retrievable token usage is charged even if the request ultimately fails due to client/server timeout. This is an exception to the page’s general non-200 statement.

Source ↗

Anthropic

Fable / Mythos 5.2
30-day retention required unless expressly authorized

Claude Fable 5.2 and Claude Mythos 5.2

Not available under ZDR unless Anthropic expressly authorizes it. Do not generalize to every Claude model.

Source ↗

Bring one representative request.

Compare the exact model/version, input, reasoning settings, output cap and required tools. Check your accepted response schema and error handling. We have not run a cross-provider compatibility or output-fidelity test.

Read the dated changes →