Changing the model name takes a minute. Knowing your JSON, tool calls and answers still hold up takes testing. We replay your real prompts on the replacement and show you exactly what changed.
Until the date OpenAI lists on its deprecations page (counted to 00:00 UTC; OpenAI doesn't give a time of day).
Three steps. You stay in control of your code and your key.
Paste code or a public GitHub repo. The checker runs in your browser and lists every retiring model ID, its shutdown date and the vendor's suggested replacement.
Send 30–200 real prompt/response pairs. We run them on the replacement model with a key you create for the job, then compare JSON validity, tool calls, refusals, length and cost.
Read the report and fix it yourself, or have us update model IDs, prompts and schemas, re-run the replay, and hand you a pull request to review.
Same prices, same delivery times. Dates come from each vendor's own deprecation page, last checked 2026-11-05.
Claude 3 Haiku, Claude 3 Sonnet, Claude 3.5 Sonnet (v1 and v2) and Claude 3.7 Sonnet reach end-of-life in several AWS regions (incl. ap-south-1, ap-southeast-1, ap-northeast-2, eu-west-2, sa-east-1). Requests fail after EOL and AWS does not migrate you automatically. In ca-central-1, Claude 3 Haiku and Sonnet stop Nov 27. Bedrock dates differ from Anthropic's own. Source
Agent Builder, the Evals dashboard/API (read-only since Oct 31) and the v1/prompts reusable-prompt objects shut down. Add-on to either package: we move agents to the Agents SDK, evals to Promptfoo, and saved prompts inline into your code. Source
claude-sonnet-4-5-20250929 retires. Replacement: claude-sonnet-5-5. Source
Requests are automatically re-routed: gemini-2.5-pro → gemini-3.8-flash, gemini-2.5-flash → gemini-3.5-flash-lite. Google says the new models have breaking API changes and are billed at their own rates. Source
gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest shut down. Replacements: gpt-image-2.5-sunburst or -flare. Source
gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3, o3-pro shut down. Replacements: the gpt-5.6 sol, terra and luna models. Source
Retires. Replacement: gpt-5.1 2025-11-13. Provisioned deployments are not upgraded automatically. Source
One-time prices, same for every model we cover. Pay with the Buy button; we email you the intake checklist right after, and your delivery clock starts once we have your files. Model usage during replays bills to your own API account, or to ours at cost, with no markup.
See what's affected.
See what breaks before your users do.
After you pay: what to send
Questions? [email protected]
We do the migration, you review it.
Questions? [email protected]
Taken from the vendors' own pages, last checked 2026-11-05.
Claude 3 Haiku / 3 Sonnet / 3.5 Sonnet (v1, v2) / 3.7 Sonnet EOL in several regions (ca-central-1: Nov 27)
Agent Builder, Evals dashboard/API, v1/prompts shut down
claude-sonnet-4-5-20250929 retires → claude-sonnet-5-5
gemini-2.5-pro / gemini-2.5-flash auto-reroute to new models; gpt-image-1-mini, gpt-image-1.5, chatgpt-image-latest shut down
gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3, o3-pro → gpt-5.6-sol / -terra / -luna
No new fine-tuning jobs for existing customers. tts-1, tts-1-hd, gpt-4o-mini-tts shut down.
GPT-4o, GPT-4o mini, GPT-4.1, GPT-4.1 mini, and fine-tunes built on them → gpt-6-luna. Azure OpenAI gpt-4o 2024-05-13 retires the same day.
whisper-1, gpt-4o-transcribe and related transcription models
Sources: OpenAI deprecations · Anthropic model deprecations · Gemini deprecations · Azure OpenAI retirements · Bedrock lifecycle
Straight answers.
Often, yes, and the free checker shows you where. The risk is behavior: a new model may format JSON differently, pick different tool arguments, or refuse things the old one answered. The replay is how you find out on your own traffic.
No. Pasted code is scanned in your browser. For repo URLs, your browser reads public files straight from GitHub.
10–200 prompt/response pairs (JSONL, CSV or a log export we can convert), plus an API key you create just for this job with a spending limit, or we run on our key and bill usage at cost. Revoke your key when we're done. Full list on the intake page.
Please remove it before sending. We'll send a redaction checklist. We delete your samples 30 days after delivery.
No. A replay shows how the new model handles the prompts you sent. It can't cover every possible input. If we find nothing worth fixing, the report says so.
No. Northgate Labs is independent.