GPT-6.1 Terra vs Claude Sonnet 5.5: API pricing for coding and agents
Terra has the lower base rates, so Sonnet has to win on results, caching or prompt length. Both have Batch API prices for work that can wait.
Changing vendors costs more than changing tiers, so the price table is where the comparison starts and not where it ends. GPT-6.1 Terra costs 50% less than Claude Sonnet 5.5 on input and 50% less on output. Both are mid tier models that people reach for when they want solid coding and agent results without paying for the top of a lineup. The real differences show up elsewhere, in how caching bills, in what happens when prompts get long, and in whether a batch discount exists for the work you can defer. This page walks through each of those with workloads that look like real products.
Compare base rates, cache reads and writes, batch prices and long prompts for mid tier coding and agent work.
By TechCompare · Updated
OpenAI: GPT-6.1 Terra
OpenAI's own price
- Input
- $1
- Output
- $5
- Cached
- $0.05
per 1M tokens
Price checked 2026-11-07
Anthropic: Claude Sonnet 5.5
Anthropic's own price
- Input
- $2
- Output
- $10
- Cached
- $0.10
per 1M tokens
Price checked 2026-11-07
Cost Comparison
Based on 100,000 input tokens (50% cached), 5,000 output tokens, and 100 requests.Prices come live from OpenRouter at each maker's own price where it's listed. Temporary deals are left out.
Side-by-side specs
| Spec | GPT-6.1 Terra | Claude Sonnet 5.5 |
|---|---|---|
| Standard input per 1M tokens | $1.00 (better on this spec) | $2.00 |
| Standard output per 1M tokens | $5.00 (better on this spec) | $10.00 |
| Cache reads per 1M tokens | $0.05 (better on this spec) | $0.10 |
| Cache writes per 1M tokens | $1.25 (better on this spec) | $2.50 |
| 100K input + 5K output, no cache hits | $0.125 (better on this spec) | $0.25 |
| 100K input + 5K output, half the input cached | $0.0775 (better on this spec) | $0.155 |
| 300K input + 3K output, no cache hits | $0.6225 (better on this spec) | $0.63 |
| Input / output on long prompts, per 1M | $2.00 / $7.50 above 272,000 input | Standard rates |
| Context window | 1,050,000 tokens | 1,000,000 tokens |
| Batch API input / output per 1M | $0.50 / $2.50 | $1.00 / $5.00 |
How they differ
Input costs $1 per million tokens on Terra and $2 on Sonnet. Output costs $5 and $10, and cache reads cost $0.05 and $0.10. At 100,000 input and 5,000 output tokens, a call costs $0.125 and $0.25 with no cache hits, and $0.0775 and $0.155 with half the input read from cache. Caching works differently on the two sides, because a cached prefix has to be written before it can be read. Sonnet's price data lists a cache write price of $2.50 per million tokens for a five minute write, and one hour writes cost $4. Terra's data lists a cache write price of $1.25 per million tokens. The warm cache examples here count reads only, so a real session pays the write cost on top, and the first call in a session is never a cache hit. A coding agent that runs 60 turns with 120,000 input tokens each, nine tenths of it read from cache and 1,500 output tokens per turn, costs $1.494 on Terra and $2.988 on Sonnet before write costs. Terra moves the whole request to a higher rate above 272,000 input tokens, and Sonnet lists no separate long prompt rate here. A document request with 300,000 input tokens and 3,000 output tokens and no cache hits costs $0.6225 on Terra and $0.63 on Sonnet, with each model's own tier applied. The embedded chart follows live OpenRouter rates, which can depend on provider routing.
Verdict
Decide with a bake off on your own tasks, scored by checks you trust. For coding, that usually means passing tests, clean diffs and a reasonable number of tool calls per task. For agents, count the turns a model needs to finish and how often it needs a human nudge. Then price what each model actually billed. On the half cached example Terra costs 50% less per call, so Sonnet needs to save enough turns, retries or review time to cover that gap. Two things can swamp the table. One is output length, since a model that explains more bills more output tokens on the same task. The other is how well the cache hits. A prompt layout that keeps the stable part first and the changing part last can matter more than a small gap in the rates. Each vendor counts tokens with its own tokenizer, so measure usage from real responses instead of assuming the same text bills the same number of tokens on both. Use the token counter to get a first estimate for your own text.
Which should you pick?
Choose GPT-6.1 Terra
Pick Terra when it passes your coding and agent evaluations and your usage looks like what its price data rewards. That means a stable prefix that the cache can serve, a large input that fits in a 1,050,000 token window, or deferrable work that can go through the Batch API. A nightly run of 20,000 documents with 6,000 input and 800 output tokens each and no cache hits costs $100 through Terra's Batch API, against $200 at standard rates. It also fits teams already on OpenAI's SDK and tooling, where staying put avoids a second integration to maintain.
Official Terra model referenceChoose Claude Sonnet 5.5
Pick Sonnet when it finishes your tasks with fewer turns or less review, or when your stack already runs on Claude. Sonnet has Batch API prices here, so the same nightly run of 20,000 documents costs $200 through batch, against $400 at standard rates. Sonnet lists no separate long prompt rate here, which keeps large document requests simple to budget within its 1,000,000 token window. Count cache writes in an agent budget, since the write fee is part of what a long session really costs.
Anthropic's pricing page