
GPT-6.1 Terra vs Sonnet 5.5: Where the Scores Tie and Split
By Emma Reed
MyClaw Editorial
Compare in MyClaw
Compare GPT-6.1 Terra and Claude Sonnet 5.5 on your own work
Give both models the same task and compare quality, speed, and cost.
AI Takeaway
- Does GPT-6.1 Terra or Sonnet 5.5 score higher? They tie at low (36) and medium (41). Sonnet then leads, 52 to 45 at xhigh (Artificial Analysis, measured Oct. 27, 2026).
- How much does the tie cost? Terra costs $0.05 a task against Sonnet's $0.35 at low, and $0.10 against $0.48 at medium (Artificial Analysis, measured Oct. 27, 2026).
- Which one answers sooner? Terra's first token comes 1.7 to 3.5x sooner from medium up. Sonnet streams about 1.2x faster (Artificial Analysis, measured Oct. 27, 2026).
- What justifies paying for Sonnet 5.5? Scores past 48, which Terra never reaches. Sonnet xhigh scores 52 at $2.01 per task (Artificial Analysis, measured Oct. 27, 2026).
- Where can I try Sonnet 5.5 on real tasks? Sonnet 5.5 runs on MyClaw agents today. Set it on a recurring task and review the output.
GPT-6.1 Terra vs Sonnet 5.5: Specs and Rates at a Glance
Sonnet 5.5 launched September 28, 2026, and Terra on October 26. Vendor-stated figures, from each company's docs and launch post (checked October 27, 2026).
- Shared: a context window near 1M tokens (Terra 1.05M, Sonnet 1M), text and image input, half-price Batch, and a 128K-token output limit.
- Per-token prices: Terra lists input at $1 and output at $5 per million tokens. Sonnet lists $2 and $10, twice the rate on both sides.
- Caching: Terra reads from cache at $0.05 and Sonnet at $0.10. Writes cost $1.25 on Terra. On Sonnet they cost $2.50 for five minutes or $4 for an hour.
- Effort: Terra has six settings, none to max, and defaults to medium. Sonnet has five, low to max, and its API default is high.
- Long prompts: over 272K tokens, Terra reprices the entire request, with input at 2x and output at 1.5x. Sonnet holds one rate through its 1M window.

Example (our arithmetic): 50K tokens in and 5K out costs $0.075 on Terra and $0.15 on Sonnet.
Solo write-ups: the Sonnet 5.5 review and the Terra review.
GPT-6.1 Terra vs Sonnet 5.5: Index, Spend and First Token by Setting
Artificial Analysis (AA) scores each effort setting on its own, so every Terra level has a Sonnet counterpart. The table shows score, spend and the wait before the first word appears.
Independent (Artificial Analysis, checked October 27, 2026):
| Effort | Intelligence Index (Terra, Sonnet) | Spend per task (Terra, Sonnet) | First token (Terra, Sonnet) |
|---|---|---|---|
| Low | 36 / 36 | $0.05 / $0.35 | 1.29s / 1.26s |
| Medium | 41 / 41 | $0.10 / $0.48 | 1.43s / 3.07s |
| High | 44 / 47 | $0.13 / $0.88 | 6.78s / 12.15s |
| Xhigh | 45 / 52 | $0.19 / $2.01 | 17.50s / 30.41s |
| Max | 48 / 56 | $0.40 / $5.46 | 131.81s / 461.94s |
AA averages spend across its own test tasks; expect your bill to differ. Its Sonnet runs use a default fallback, which means some requests can end on another model. The tie rests on rounded values: Sonnet scores 35.9 at low and 40.8 at medium.
Sonnet's lead is 0 points at low and medium, 3 at high, 7 at xhigh and 8 at max (our arithmetic). At the two tied levels, Sonnet spends 7.0x and 4.8x as much.
For Terra against Sol, see Terra and Sol side by side.
GPT-6.1 Terra vs Sonnet 5.5: Where the Score Gap Opens
Three pairings across effort levels show the trade (our arithmetic).
- Terra high vs Sonnet medium. Terra scores 44 to 41 and spends $0.13 to $0.48. Sonnet medium answers first, in 3.07s to 6.78s.
- Terra max vs Sonnet high. Terra scores 48 to 47 and spends $0.40 to $0.88. Terra max's first token takes 131.81s. Sonnet high's takes 12.15s.
- Sonnet xhigh vs Terra max. Sonnet scores 52 to 48 and starts in 30.41s against 131.81s. It spends $2.01 against $0.40.
Independent (Artificial Analysis, checked October 27, 2026), at max effort: GDPval-AA reads 48.8% for Terra and 66.9% for Sonnet. Terminal-Bench 4.0 reads 55.6% to 63.6%. AA-Omniscience accuracy reads 49.1% to 53.9%. Terra posts the better non-hallucination rate, 62.0% to 53.0%.
Vendors report Terminal-Bench 4.0 differently. OpenAI states 50.3% for Terra at max (OpenAI, stated October 26, 2026). Anthropic states 70.6% for Sonnet 5.5 (Anthropic, stated September 28, 2026). Those figures leave a 20-point gap. AA's own runs leave 8. Compare scores only within one source.
If Sonnet max still falls short, read Sonnet 5.5 against Opus 5.5 by effort.

GPT-6.1 Terra vs Sonnet 5.5 Speed: Testing the Vendor Labels
Anthropic's docs label Sonnet 5.5 latency "Fast." Its launch post says output runs at least 30% quicker than Sonnet 5 (Anthropic, stated September 28, 2026). OpenAI calls Terra much quicker, but its 3x figure is measured against Astra and Sol. It says nothing about Sonnet.
Independent (Artificial Analysis, checked October 27, 2026):
- Wait for a first token: level at low, 1.29s for Terra and 1.26s for Sonnet. Terra is then sooner by 2.1x at medium, 1.8x at high, 1.7x at xhigh and 3.5x at max (our arithmetic).
- Output speed: Terra streams near 91 tokens per second whatever the effort. Sonnet runs 106.5 to 109.9 up to xhigh, then 141.2 at max.
Third-party (OpenRouter, checked October 27, 2026): Terra's single OpenAI endpoint shows 2.53s latency and 64 tokens per second. Across Sonnet 5.5's five standard providers, latency runs 1.14s to 3.59s at 90 to 125 tokens per second.
"Fast" holds for streaming. Terra wins the first-word wait from medium up. Live routing narrows both gaps, so time your own runs.
GPT-6.1 Terra vs Sonnet 5.5 Cost: The Cheapest Route to Each Score
OpenAI says Terra advances the "cost-intelligence frontier" (OpenAI, stated October 26, 2026). Each line shows the cheapest AA entry that reaches the score (our arithmetic):
- Index 36: Terra low, $0.05.
- Index 41: Terra medium, $0.10.
- Index 44: Terra high, $0.13.
- Index 47 to 48: Terra max, $0.40. Sonnet high reaches 47 at $0.88.
- Index 52: Sonnet xhigh only, $2.01.
- Index 56: Sonnet max only, $5.46.
The claim holds up to Index 48. Past that, Terra has no entry.
Sonnet also writes more. AA counts 23M output tokens at low against Terra's 7.9M, and 420M at max against 130M. That is 2.2x to 3.2x as many (AA data; ratios are our arithmetic).
Rates explain part of the spend gap. Twice the rate times the token ratio gives 4.4x at medium (AA: 4.8x) but 6.5x at max (AA: 13.7x). AA does not show why.
GPT-6.1 Terra vs Sonnet 5.5: Routing Rules for Everyday Work
- Low or medium on Terra: volume work and screen tasks. OpenAI points Terra at forms, screenshots and desktop steps (OpenAI, stated October 26, 2026). At equal score it spends 4.8x to 7.0x less.
- Terra high: when 44 is enough, at $0.13 a task.
- Sonnet high or xhigh: scores of 47 to 52, with a first token in 12s to 30s. Terra max reaches 48 but needs 131.81s.
- Max on either: jobs that can wait. Sonnet max needs 461.94s before its first token.
- Prompts over 272K tokens: Terra reprices every token in the request and Sonnet does not, so the spend gap narrows (vendor-stated).
For pricier rivals, see Sonnet 5.5 set against GPT-6 Sol and how Terra measures up to Opus 5.5.
Run Sonnet 5.5 for Real Work on a MyClaw Agent

Averages give a first guess. Your recurring tasks give the real answer.
On MyClaw, OpenClaw and Hermes agents are set up and kept running for you. They work through scheduled jobs around the clock, even when you are offline. You maintain no servers. Any agent you deploy works with most popular models. Sonnet 5.5 is one you can pick today.
- Take a daily report or inbox triage as your test job.
- Run it on Sonnet 5.5 at medium and on the model you use now, with the same prompt.
- After a week, compare how many answers you kept, how many runs needed a retry and how long each took.
Then keep whichever setup meets your bar.
GPT-6.1 Terra vs Sonnet 5.5 Verdict
- Terra low or medium when that score is enough. Same Index, 4.8x to 7.0x less spend.
- Terra high for a score near 44.
- Sonnet 5.5 high or above for 47 and up.
- Max on either for work that can wait.
Name the score your job needs. Then take the first rung of the ladder that reaches it.
Compare in MyClaw
Compare GPT-6.1 Terra and Claude Sonnet 5.5 on your own work
Give both models the same task and compare quality, speed, and cost.