| GPT-5.6 Terra | GPT-5.5 | |
|---|---|---|
| Price in/out ($/M) | $1 / $6 | $5 / $30 |
Terra undercuts GPT-5.5 hard, $1 / $6 per M tokens against $5 / $30, five times cheaper on both sides, and it brings a 1.05M context window against 600K. The catch is the same one that bites the whole GPT-5.6 line: GPT-5.5 takes text, vision and audio, and Terra only takes text and vision. If any part of your product speaks or listens, this is not a migration, it is a rewrite. The second catch is that OpenAI publishes no benchmarks for Terra, so the numbers you can compare are price, context and modality, and nothing else. Where GPT-5.5 is measured, it is strong: 66% on SWE-bench Verified, 92% on AIME, 1432 Arena Elo. For high volume text and vision work Terra is the obvious cost move, but benchmark it on your own workload first, because nobody else has.