---
title: "Migration guide: retiring models and like-for-like replacements · Undominated.ai"
canonical: https://undominated.ai/migrate/
description: "27 models with a retirement date as of 2026-10-06, each with the replacement that loses no recorded capability, scores at least as high where the model is rated, and is cheapest at short prompts, or a plain statement that none exists."
---

# Migration guide: retiring models and like-for-like replacements · Undominated.ai

> 27 models with a retirement date as of 2026-10-06, each with the replacement that loses no recorded capability, scores at least as high where the model is rated, and is cheapest at short prompts, or a plain statement that none exists.

# Migration guide

Every model with a standard listing and a retirement date, one whose date has passed but which is still served included, and its like-for-like replacement: of the models with no retirement date that keep every recorded capability and, when the retiring model is rated, score at least as high, the one that is cheapest at short prompts. Where none exists, the page says so.

[Retirement tracking](/deprecations/) [Methodology](/methodology/)

As of 2026-10-06 , 27 models carry a retirement date. 27 of them have a like-for-like replacement . 17 of the 27 retiring models are unrated, so their replacements are chosen on recorded capabilities and price, not score.

**Like-for-like** means every input and output modality, tool use and reasoning the retiring model has, a context window and output limit at least as large and on record, and no retirement date of its own. It covers what the catalogue records, not every behaviour. Price is the balanced workload, 3 parts input to 1 part output, in $ per million tokens, for a prompt below every context tier: "cheapest" means cheapest at short prompts. Where a longer prompt makes another model the cheapest (10 of 27 lines), the line names it; how prices rise with prompt size is on [Context price cliffs](/cliffs/). Scores are LMArena Elo under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/), from [the official dataset](https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset).

 - [Nano Banana (Gemini 2.5 Flash Image)](/models/google__gemini-2.5-flash-image/) → replaced by [Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)](/models/google__gemini-3.1-flash-lite-image/): Neither is rated, so this compares recorded capabilities and price, not score; 34% cheaper Retirement date 2026-10-02 · score unrated → unrated · $0.850 → $0.563 per million
- [ERNIE 4.5 VL 424B A47B](/models/baidu__ernie-4.5-vl-424b-a47b/) → replaced by [Nex-N2.5-Mini](/models/nex-agi__nex-n2.5-mini/): Neither is rated, so this compares recorded capabilities and price, not score; 93% cheaper Retirement date 2026-10-08 · score unrated → unrated · $0.628 → $0.044 per million
- [Qwen Plus 0728](/models/qwen__qwen-plus-2025-07-28/) → replaced by [Qwen3.7 Flash](/models/qwen__qwen3.7-flash/): Neither is rated, so this compares recorded capabilities and price, not score; 86% cheaper Over 32K-token prompts, [DeepSeek V4 Flash 0731](/models/deepseek__deepseek-v4-flash-0731/) is the cheapest instead. Retirement date 2026-10-09 · score unrated → unrated · $0.390 → $0.055 per million
- [Qwen3 235B A22B](/models/qwen__qwen3-235b-a22b/) → replaced by [DeepSeek V4 Flash 0423](/models/deepseek__deepseek-v4-flash/): +65.5 score, 86% cheaper Retirement date 2026-10-09 · score 1366.6 → 1432.1 · $0.796 → $0.113 per million
- [Qwen3 235B A22B Thinking 2507](/models/qwen__qwen3-235b-a22b-thinking-2507/) → replaced by [DeepSeek V4 Flash 0423](/models/deepseek__deepseek-v4-flash/): +16.9 score, 85% cheaper Retirement date 2026-10-09 · score 1415.2 → 1432.1 · $0.748 → $0.113 per million
- [Qwen3 30B A3B Thinking 2507](/models/qwen__qwen3-30b-a3b-thinking-2507/) → replaced by [gpt-oss-20b](/models/openai__gpt-oss-20b/): Qwen3 30B A3B Thinking 2507 is unrated, so this compares recorded capabilities and price, not score; 95% cheaper Retirement date 2026-10-09 · score unrated → 1287.7 · $0.750 → $0.036 per million
- [Qwen3 8B](/models/qwen__qwen3-8b/) → replaced by [gpt-oss-20b](/models/openai__gpt-oss-20b/): Qwen3 8B is unrated, so this compares recorded capabilities and price, not score; 82% cheaper Retirement date 2026-10-09 · score unrated → 1287.7 · $0.202 → $0.036 per million
- [Qwen3 Coder Plus](/models/qwen__qwen3-coder-plus/) → replaced by [Qwen3.7 Flash](/models/qwen__qwen3.7-flash/): Neither is rated, so this compares recorded capabilities and price, not score; 96% cheaper Over 32K-token prompts, [DeepSeek V4 Flash 0731](/models/deepseek__deepseek-v4-flash-0731/) is the cheapest instead. Retirement date 2026-10-09 · score unrated → unrated · $1.30 → $0.055 per million
- [Qwen3 Max](/models/qwen__qwen3-max/) → replaced by [DeepSeek V4 Flash 0423](/models/deepseek__deepseek-v4-flash/): +19.7 score, 93% cheaper Retirement date 2026-10-09 · score 1412.4 → 1432.1 · $1.56 → $0.113 per million
- [Qwen3 Max Thinking](/models/qwen__qwen3-max-thinking/) → replaced by [Qwen3.7 Flash](/models/qwen__qwen3.7-flash/): Neither is rated, so this compares recorded capabilities and price, not score; 96% cheaper Over 32K-token prompts, [Nemotron 3 Nano 30B A3B](/models/nvidia__nemotron-3-nano-30b-a3b/) is the cheapest instead. Retirement date 2026-10-09 · score unrated → unrated · $1.56 → $0.055 per million
- [Qwen3 Next 80B A3B Thinking](/models/qwen__qwen3-next-80b-a3b-thinking/) → replaced by [DeepSeek V4 Flash 0423](/models/deepseek__deepseek-v4-flash/): +63.8 score, 73% cheaper Retirement date 2026-10-09 · score 1368.3 → 1432.1 · $0.413 → $0.113 per million
- [Qwen3 VL 235B A22B Thinking](/models/qwen__qwen3-vl-235b-a22b-thinking/) → replaced by [Gemma 4 26B A4B](/models/google__gemma-4-26b-a4b-it/): +33.2 score, 90% cheaper Retirement date 2026-10-09 · score 1400.7 → 1433.9 · $1.30 → $0.127 per million
- [Qwen3 VL 30B A3B Thinking](/models/qwen__qwen3-vl-30b-a3b-thinking/) → replaced by [Qwen3.7 Flash](/models/qwen__qwen3.7-flash/): Neither is rated, so this compares recorded capabilities and price, not score; 88% cheaper Over 32K-token prompts, [Ling 3.0 Flash VL](/models/inclusionai__ling-3.0-flash-vl/) is the cheapest instead. Retirement date 2026-10-09 · score unrated → unrated · $0.468 → $0.055 per million
- [Qwen3 VL 32B Instruct](/models/qwen__qwen3-vl-32b-instruct/) → replaced by [Qwen3.7 Flash](/models/qwen__qwen3.7-flash/): Neither is rated, so this compares recorded capabilities and price, not score; 70% cheaper Over 32K-token prompts, [Ling 3.0 Flash VL](/models/inclusionai__ling-3.0-flash-vl/) is the cheapest instead. Retirement date 2026-10-09 · score unrated → unrated · $0.182 → $0.055 per million
- [Qwen3 VL 8B Instruct](/models/qwen__qwen3-vl-8b-instruct/) → replaced by [Qwen3.7 Flash](/models/qwen__qwen3.7-flash/): Neither is rated, so this compares recorded capabilities and price, not score; 73% cheaper Over 32K-token prompts, [Ling 3.0 Flash VL](/models/inclusionai__ling-3.0-flash-vl/) is the cheapest instead. Retirement date 2026-10-09 · score unrated → unrated · $0.202 → $0.055 per million
- [Qwen3 VL 8B Thinking](/models/qwen__qwen3-vl-8b-thinking/) → replaced by [Qwen3.7 Flash](/models/qwen__qwen3.7-flash/): Neither is rated, so this compares recorded capabilities and price, not score; 92% cheaper Over 32K-token prompts, [Ling 3.0 Flash VL](/models/inclusionai__ling-3.0-flash-vl/) is the cheapest instead. Retirement date 2026-10-09 · score unrated → unrated · $0.660 → $0.055 per million
- [Qwen3.6 Max Preview](/models/qwen__qwen3.6-max-preview/) → replaced by [MiMo-V2.6-Flash](/models/xiaomi__mimo-v2.6-flash/): +9.6 score, 92% cheaper Retirement date 2026-10-09 · score 1446.8 → 1456.4 · $2.31 → $0.175 per million
- [Gemini 2.5 Flash](/models/google__gemini-2.5-flash/) → replaced by [Gemini 3.5 Flash Lite](/models/google__gemini-3.5-flash-lite/): +17.7 score, same price Retirement date 2026-10-20 · score 1417.0 → 1434.7 · $0.850 → $0.850 per million
- [Gemini 2.5 Flash Lite](/models/google__gemini-2.5-flash-lite/) → replaced by [Muse Spark 1.2 Contributor](/models/meta__muse-spark-1.2-contributor/): Neither is rated, so this compares recorded capabilities and price, not score; 29% cheaper Retirement date 2026-10-20 · score unrated → unrated · $0.175 → $0.125 per million
- [Gemini 2.5 Pro](/models/google__gemini-2.5-pro/) → replaced by [Gemini 3.6 Flash](/models/google__gemini-3.6-flash/): +21.7 score, 56% cheaper Retirement date 2026-10-20 · score 1457.8 → 1479.5 · $3.44 → $1.50 per million
- [Laguna S 2.1](/models/poolside__laguna-s-2.1/) → replaced by [DeepSeek V4 Flash 0731](/models/deepseek__deepseek-v4-flash-0731/): Neither is rated, so this compares recorded capabilities and price, not score; 28% cheaper Retirement date 2026-10-31 · score unrated → unrated · $0.125 → $0.090 per million
- [Laguna XS 2.1](/models/poolside__laguna-xs-2.1/) → replaced by [Qwen3.7 Flash](/models/qwen__qwen3.7-flash/): Neither is rated, so this compares recorded capabilities and price, not score; 56% cheaper Over 32K-token prompts, [Nemotron 3.5 Lightning](/models/nvidia__nemotron-3.5-lightning/) is the cheapest instead. Retirement date 2026-10-31 · score unrated → unrated · $0.125 → $0.055 per million
- [Seed 1.6](/models/bytedance-seed__seed-1.6/) → replaced by [Qwen3.7 Flash](/models/qwen__qwen3.7-flash/): Neither is rated, so this compares recorded capabilities and price, not score; 92% cheaper Over 32K-token prompts, [Ling 3.0 Flash VL](/models/inclusionai__ling-3.0-flash-vl/) is the cheapest instead. Retirement date 2026-11-11 · score unrated → unrated · $0.688 → $0.055 per million
- [Seed 1.6 Flash](/models/bytedance-seed__seed-1.6-flash/) → replaced by [Qwen3.7 Flash](/models/qwen__qwen3.7-flash/): Neither is rated, so this compares recorded capabilities and price, not score; 58% cheaper Over 32K-token prompts, [Ling 3.0 Flash VL](/models/inclusionai__ling-3.0-flash-vl/) is the cheapest instead. Retirement date 2026-11-11 · score unrated → unrated · $0.131 → $0.055 per million
- [Seed-2.0-Code](/models/bytedance-seed__seed-2.0-code/) → replaced by [Muse Spark 1.2 Contributor](/models/meta__muse-spark-1.2-contributor/): Neither is rated, so this compares recorded capabilities and price, not score; 89% cheaper Retirement date 2026-11-11 · score unrated → unrated · $1.13 → $0.125 per million
- [GLM 4.5](/models/z-ai__glm-4.5/) → replaced by [DeepSeek V4 Flash 0423](/models/deepseek__deepseek-v4-flash/): +1.8 score, 89% cheaper Retirement date 2026-12-31 · score 1430.3 → 1432.1 · $1.00 → $0.113 per million
- [GLM 4.7](/models/z-ai__glm-4.7/) → replaced by [MiMo-V2.6-Flash](/models/xiaomi__mimo-v2.6-flash/): +20.9 score, 76% cheaper Retirement date 2026-12-31 · score 1435.5 → 1456.4 · $0.738 → $0.175 per million

Dates are the catalogue's, as of 2026-10-06. Direct API and partner schedules can differ; check the notice on [Retirement tracking](/deprecations/) before migrating.