Meta Llama 3.2, Llama 3.3 & Llama 4 Deprecation, Shutdown Dates & Migration Guide
Runs locally in your browser — 0 bytes uploaded
Synced 2026-11-12 UTC · 12 variants · 2 ecosystems
Track confirmed UTC deprecation and shutdown dates for Meta Llama 3.2, Llama 3.3 & Llama 4 (Meta) across AWS Bedrock, Azure OpenAI. Preview meta.llama3-2-90b-instruct-v1:0 → meta.llama4-maverick-17b-instruct-v1:0 upgrades locally below, audit full configs in the Model Config Auditor, or benchmark token pricing in DevToken.
Tracked Variants
12
Earliest Upcoming EOL
2026-11-19
Recommended Upgrade
meta.llama4-maverick-17b-instruct-v1:0
Ecosystems Covered
AWS Bedrock, Azure OpenAI
1. Quick .env & SDK Snippet Replacement Matcher
Edit or paste your .env assignment, LiteLLM route, or SDK model identifier below to check live 3-state EOL status and copy a ready-to-apply replacement diff locally in your browser.
Replacement Diff Preview (meta.llama3-2-90b-instruct-v1:0 → meta.llama4-maverick-17b-instruct-v1:0)
# BEDROCK_MODEL_ID: meta.llama3-2-90b-instruct-v1:0 -> meta.llama4-maverick-17b-instruct-v1:0 [RETIRED · EOL 2026-07-07]
- BEDROCK_MODEL_ID="meta.llama3-2-90b-instruct-v1:0"
+ BEDROCK_MODEL_ID="meta.llama4-maverick-17b-instruct-v1:0"
# AZURE_OPENAI_DEPLOYMENT: Llama-4-Maverick-17B-128E-Instruct-FP8 -> meta.llama4-maverick-17b-instruct-v1:0 [DEPRECATED · EOL 2026-11-19]
- AZURE_OPENAI_DEPLOYMENT="Llama-4-Maverick-17B-128E-Instruct-FP8"
+ AZURE_OPENAI_DEPLOYMENT="meta.llama4-maverick-17b-instruct-v1:0"2. Cross-Ecosystem Meta Llama 3.2, Llama 3.3 & Llama 4 Lifecycle Matrix (12)
Direct vendor APIs and cloud platforms retire dated snapshots on independent schedules. Verify the exact model ID, vendor lifecycle stage, and UTC shutdown date matching your deployment.
| Model ID | Provider | Calculated Status | Confirmed Shutdown (UTC) | Vendor Lifecycle | Replacement & Migration | Source |
|---|---|---|---|---|---|---|
Llama-4-Maverick-17B-128E-Instruct-FP8 |
Azure OpenAI | DEPRECATED (7d) | 2026-11-19in 7d | GA | Docs ↗ | |
Llama-3.3-70B-Instruct |
Azure OpenAI | DEPRECATED (80d) | 2027-01-31in 80d | GA | Docs ↗ | |
Llama-4-Scout-17B-16E-Instruct |
Azure OpenAI | DEPRECATED (80d) | 2027-01-31in 80d | GA | Docs ↗ | |
meta.llama3-3-70b-instruct-v1:0 |
AWS Bedrock(Meta) | DEPRECATED (138d) | 2027-03-30in 138d | LegacyDep: 2026-09-30us-east-1, us-east-2, us-west-2 | Docs ↗ | |
meta.llama4-maverick-17b-instruct-v1:0 |
AWS Bedrock(Meta) | DEPRECATED (138d) | 2027-03-30in 138d | LegacyDep: 2026-09-30us-east-1, us-east-2, us-west-1, us-west-2 | Docs ↗ | |
meta.llama4-scout-17b-instruct-v1:0 |
AWS Bedrock(Meta) | DEPRECATED (138d) | 2027-03-30in 138d | LegacyDep: 2026-09-30us-east-1, us-east-2, us-west-1, us-west-2 | Docs ↗ | |
meta.llama3-2-11b-instruct-v1:0 |
AWS Bedrock(Meta) | RETIRED | 2026-07-07128d ago | EOLDep: 2026-01-07us-east-1, us-east-2, us-west-2 | Docs ↗ | |
meta.llama3-2-1b-instruct-v1:0 |
AWS Bedrock(Meta) | RETIRED | 2026-07-07128d ago | EOLDep: 2026-01-07eu-central-1, eu-west-1, eu-west-3, us-east-1, us-east-2, us-west-2 | Docs ↗ | |
meta.llama3-2-3b-instruct-v1:0 |
AWS Bedrock(Meta) | RETIRED | 2026-07-07128d ago | EOLDep: 2026-01-07eu-central-1, eu-west-1, eu-west-3, us-east-1, us-east-2, us-west-2 | Docs ↗ | |
meta.llama3-2-90b-instruct-v1:0 |
AWS Bedrock(Meta) | RETIRED | 2026-07-07128d ago | EOLDep: 2026-01-07us-east-1, us-east-2, us-west-2 | Docs ↗ | |
Llama-3.2-11B-Vision-Instruct |
Azure OpenAI | RETIRED | 2026-06-13152d ago | Retired | Docs ↗ | |
Llama-3.2-90B-Vision-Instruct |
Azure OpenAI | RETIRED | 2026-06-13152d ago | Retired | Docs ↗ |
AWS Bedrock regional scope applies across 7 tracked regions (eu-central-1, eu-west-1, eu-west-3, us-east-1, us-east-2, us-west-1); verify regional inference prefixes (us., eu., apac., global.) before cutover.
3. Production Migration Checklist (meta.llama3-2-90b-instruct-v1:0 → meta.llama4-maverick-17b-instruct-v1:0)
Follow this deterministic engineering sequence before retiring pinned snapshots in production services and SDK clients:
- Inventory pinned model IDs: Search repositories and deployment manifests for
BEDROCK_MODEL_ID,bedrock/LiteLLM routes, and dated snapshot strings in the Meta Llama 3.2, Llama 3.3 & Llama 4 family (6 deprecated, 6 retired, 0 active variants tracked). - Benchmark successor quality & token pricing: Validate structured output schemas, tool-calling payloads, and latency against
meta.llama4-maverick-17b-instruct-v1:0, then compare input, output, and cached token rates in DevToken. - Enforce keyless CI guardrails: Wire RSS, Atom, iCal feeds & CI Guard into pull-request checks so retired or deprecated Meta Llama 3.2, Llama 3.3 & Llama 4 identifiers cannot re-enter production.
4. Related Meta & Ecosystem Migration Guides
Compare sibling model families or inspect full provider-level schedules across every ecosystem hosting this family:
Mistral Large, Small, Ministral & Codestral
DEPRECATED (27d)Cohere Command R, Command R+ & Command A
DEPRECATED (95d)GPT-4o mini
DEPRECATED (63d)GPT-4o
DEPRECATED (63d)5. Meta Llama 3.2, Llama 3.3 & Llama 4 Deprecation & Upgrade FAQ
Is Meta Llama 3.2, Llama 3.3 & Llama 4 deprecated or retired across AWS Bedrock, Azure OpenAI?
LLMEOL tracks 12 Meta Llama 3.2, Llama 3.3 & Llama 4 variants across AWS Bedrock, Azure OpenAI: 6 deprecated with scheduled shutdowns, 6 retired past their UTC EOL date, and 0 active with no announced shutdown date. Check the cross-ecosystem matrix above for exact snapshot dates.
What is the recommended replacement model when migrating off meta.llama3-2-90b-instruct-v1:0?
The primary recommended upgrade target for meta.llama3-2-90b-instruct-v1:0 is meta.llama4-maverick-17b-instruct-v1:0. Always test prompt formatting, max output tokens, and tool schemas on meta.llama4-maverick-17b-instruct-v1:0, and compare token pricing in DevToken before switching production traffic.
How can I replace meta.llama3-2-90b-instruct-v1:0 in .env and LiteLLM configs safely?
Paste your BEDROCK_MODEL_ID assignment or LiteLLM model block into the Quick Replacement Matcher on this page to generate an exact diff, or open the full Local Model Config Auditor to scan multi-service manifests with zero network upload.