Axis Spatial Spatial Intelligence Index

Search modelsMenu

OpenEarthAgent Sentinel-2 benchmark leaderboard

Land change from Sentinel-2 index rasters, from OpenEarthAgent Sentinel-2. A standalone evaluation: its scores are shown here and on model pages, but it is not counted in any index or ranking until more models have run it.

Score

All

OpenEarthAgent Sentinel-2 score

Share of tasks passed, 0 to 100 · Higher is better

OpenEarthAgent Sentinel-2 scoreScores 0 to 100: Nemotron 3 Ultra (high), Terminus-2: 100; Step 5 Preview (Reasoning), Terminus-2: 100; Claude Haiku 4.5 (Reasoning), Claude Code: 100; Claude Opus 5.5 (medium), Claude Code: 100; Claude Opus 5.5 (high), Claude Code: 100; Claude Sonnet 5.5 (medium), Claude Code: 100; Claude Sonnet 5.5 (high), Claude Code: 100; Claude Haiku 5.5 (medium), Claude Code: 100; Claude Haiku 5.5 (high), Claude Code: 100; Claude Opus 5.5 (xhigh), Claude Code: 100; Claude Sonnet 5.5 (xhigh), Claude Code: 100; Claude Haiku 5.5 (xhigh), Claude Code: 100; Claude Opus 5.5 (low), Claude Code: 100; Claude Sonnet 5.5 (low), Claude Code: 100; Claude Haiku 5.5 (low), Claude Code: 100; Claude Opus 5.5 (max), Claude Code: 100; Claude Haiku 5.5 (max), Claude Code: 100; Claude Opus 5 (high), Claude Code: 100; Claude Sonnet 5 (high), Claude Code: 100; Claude Opus 4.8 (high), Claude Code: 100; Claude Opus 4.7 (high), Claude Code: 100; Mistral Large 4 (high), Terminus-2: 99; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 97; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 93; Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2: 520255075100Nemotron 3 Ultra (high)Terminus-2100Step 5 Preview (Reasoning)Terminus-2100Claude Haiku 4.5 (Reasoning)Claude Code100Claude Opus 5.5 (medium)Claude Code100Claude Opus 5.5 (high)Claude Code100Claude Sonnet 5.5 (medium)Claude Code100Claude Sonnet 5.5 (high)Claude Code100Claude Haiku 5.5 (medium)Claude Code100Claude Haiku 5.5 (high)Claude Code100Claude Opus 5.5 (xhigh)Claude Code100Claude Sonnet 5.5 (xhigh)Claude Code100Claude Haiku 5.5 (xhigh)Claude Code100Claude Opus 5.5 (low)Claude Code100Claude Sonnet 5.5 (low)Claude Code100Claude Haiku 5.5 (low)Claude Code100Claude Opus 5.5 (max)Claude Code100Claude Haiku 5.5 (max)Claude Code100Claude Opus 5 (high)Claude Code100Claude Sonnet 5 (high)Claude Code100Claude Opus 4.8 (high)Claude Code100Claude Opus 4.7 (high)Claude Code100Mistral Large 4 (high)Terminus-299Nemotron 3.5 Super VL 120B (Reasoning)Terminus-297Nemotron 3.5 Lightning 30B (Reasoning)Terminus-293Nemotron 3 Nano Omni 30B (Reasoning)Terminus-252
How to read this chart

What this metric means. The share of tasks the entry passed, 0 to 100.

Data table: OpenEarthAgent Sentinel-2 score

Open weights

OpenEarthAgent Sentinel-2 score

Share of tasks passed, 0 to 100 · Higher is better

OpenEarthAgent Sentinel-2 scoreScores 0 to 100: Nemotron 3 Ultra (high), Terminus-2: 100; Step 5 Preview (Reasoning), Terminus-2: 100; Mistral Large 4 (high), Terminus-2: 99; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 97; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 93; Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2: 520255075100Nemotron 3 Ultra (high)Terminus-2100Step 5 Preview (Reasoning)Terminus-2100Mistral Large 4 (high)Terminus-299Nemotron 3.5 Super VL 120B (Reasoning)Terminus-297Nemotron 3.5 Lightning 30B (Reasoning)Terminus-293Nemotron 3 Nano Omni 30B (Reasoning)Terminus-252
How to read this chart

What this metric means. The share of tasks the entry passed, 0 to 100.

Data table: OpenEarthAgent Sentinel-2 score

All

OpenEarthAgent Sentinel-2: score vs. cost per task

OpenEarthAgent Sentinel-2 score against average cost per task (USD) · Higher and further left is better

  • Most attractive quadrant
  • Pareto line
OpenEarthAgent Sentinel-2: score vs. cost per taskOpenEarthAgent Sentinel-2 score (0 to 100) against cost per task in US dollars, log scale: Mistral Large 4 (high), Terminus-2: 99.18, $0.013; Nemotron 3 Ultra (high), Terminus-2: 100, $0.015; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 93.44, $0.0027; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 96.72, $0.0036; Step 5 Preview (Reasoning), Terminus-2: 100, $0.025; Claude Haiku 4.5 (Reasoning), Claude Code: 100, $0.033; Claude Opus 5.5 (medium), Claude Code: 100, $0.063; Claude Opus 5.5 (high), Claude Code: 100, $0.075; Claude Sonnet 5.5 (medium), Claude Code: 100, $0.028; Claude Sonnet 5.5 (high), Claude Code: 100, $0.030; Claude Haiku 5.5 (medium), Claude Code: 100, $0.0028; Claude Haiku 5.5 (high), Claude Code: 100, $0.0036; Claude Opus 5.5 (xhigh), Claude Code: 100, $0.096; Claude Sonnet 5.5 (xhigh), Claude Code: 100, $0.040; Claude Haiku 5.5 (xhigh), Claude Code: 100, $0.0046; Claude Opus 5.5 (low), Claude Code: 100, $0.058; Claude Sonnet 5.5 (low), Claude Code: 100, $0.029; Claude Haiku 5.5 (low), Claude Code: 100, $0.0025; Claude Opus 5.5 (max), Claude Code: 100, $0.49; Claude Haiku 5.5 (max), Claude Code: 100, $0.021; Claude Opus 5 (high), Claude Code: 100, $0.14; Claude Sonnet 5 (high), Claude Code: 100, $0.058; Claude Opus 4.8 (high), Claude Code: 100, $0.093; Claude Opus 4.7 (high), Claude Code: 100, $0.13Most attractive quadrant9293949596979899100$0.002$0.005$0.01$0.02$0.05$0.10$0.20$0.50$1.00Cost per task (USD, log scale)OpenEarthAgent Sentinel-2 scoreNemotron 3 Ultra (high), Terminus-2Step 5 Preview (Reasoning), Terminus-2Claude Haiku 4.5 (Reasoning), Claude CodeClaude Haiku 5.5 (low), Claude CodeClaude Opus 5.5 (medium), Claude CodeClaude Opus 5.5 (high), Claude CodeClaude Sonnet 5.5 (medium), Claude CodeClaude Sonnet 5.5 (high), Claude CodeClaude Haiku 5.5 (medium), Claude CodeClaude Haiku 5.5 (high), Claude CodeClaude Opus 5.5 (xhigh), Claude CodeClaude Sonnet 5.5 (xhigh), Claude CodeClaude Haiku 5.5 (xhigh), Claude CodeClaude Opus 5.5 (low), Claude CodeClaude Sonnet 5.5 (low), Claude CodeClaude Opus 5.5 (max), Claude CodeClaude Haiku 5.5 (max), Claude CodeClaude Opus 5 (high), Claude CodeClaude Sonnet 5 (high), Claude CodeClaude Opus 4.8 (high), Claude CodeClaude Opus 4.7 (high), Claude CodeMistral Large 4 (high), Terminus-2Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2
How to read this chart

What cost is measuring. What one attempt at a task would cost if paid token by token at the provider's published API prices, averaged over attempts: ordinary input tokens, input tokens served from the provider's cache at the lower cached price, tokens written to the cache where the provider charges for that, and output tokens (thinking included). Cost assumes each harness uses the model's pay-per-token API, priced at public list rates; sandboxes and our own engineering time are not included. Failed attempts count too.

How to read this chart. Each point is one entry. Further left means a lower average cost per task (log scale); higher means a higher OpenEarthAgent Sentinel-2 score. Both axes fit the points shown, so the score axis does not start at zero. The shaded most attractive quadrant is the top-left quarter of the plot, split at the middle of each axis: a reading aid, not a fixed bar. The dotted Pareto line joins the points that no cheaper one beats on score. Only exact costs are plotted; the table lists every entry.

Data table: OpenEarthAgent Sentinel-2: score vs. cost per task

Open weights

OpenEarthAgent Sentinel-2: score vs. cost per task

OpenEarthAgent Sentinel-2 score against average cost per task (USD) · Higher and further left is better

  • Most attractive quadrant
  • Pareto line
OpenEarthAgent Sentinel-2: score vs. cost per taskOpenEarthAgent Sentinel-2 score (0 to 100) against cost per task in US dollars, log scale: Mistral Large 4 (high), Terminus-2: 99.18, $0.013; Nemotron 3 Ultra (high), Terminus-2: 100, $0.015; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 93.44, $0.0027; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 96.72, $0.0036; Step 5 Preview (Reasoning), Terminus-2: 100, $0.025Most attractive quadrant9293949596979899100$0.002$0.005$0.01$0.02$0.05Cost per task (USD, log scale)OpenEarthAgent Sentinel-2 scoreNemotron 3 Ultra (high), Terminus-2Mistral Large 4 (high), Terminus-2Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2Step 5 Preview (Reasoning), Terminus-2
How to read this chart

What cost is measuring. What one attempt at a task would cost if paid token by token at the provider's published API prices, averaged over attempts: ordinary input tokens, input tokens served from the provider's cache at the lower cached price, tokens written to the cache where the provider charges for that, and output tokens (thinking included). Cost assumes each harness uses the model's pay-per-token API, priced at public list rates; sandboxes and our own engineering time are not included. Failed attempts count too.

How to read this chart. Each point is one entry. Further left means a lower average cost per task (log scale); higher means a higher OpenEarthAgent Sentinel-2 score. Both axes fit the points shown, so the score axis does not start at zero. The shaded most attractive quadrant is the top-left quarter of the plot, split at the middle of each axis: a reading aid, not a fixed bar. The dotted Pareto line joins the points that no cheaper one beats on score. Only exact costs are plotted; the table lists every entry.

Data table: OpenEarthAgent Sentinel-2: score vs. cost per task
OpenEarthAgent Sentinel-2: score vs. cost per task
ModelOpenEarthAgent Sentinel-2 scoreCost per task (USD)On the Pareto line
Nemotron 3 Ultra (high), Terminus-2100$0.015Yes
Step 5 Preview (Reasoning), Terminus-2100$0.025No
Mistral Large 4 (high), Terminus-299$0.013Yes
Nemotron 3.5 Super VL 120B (Reasoning), Terminus-297$0.0036Yes
Nemotron 3.5 Lightning 30B (Reasoning), Terminus-293$0.0027Yes
Nemotron 3 Nano Omni 30B (Reasoning), Terminus-252No dataNo data

Token usage

All

OpenEarthAgent Sentinel-2: tokens per task

Average tokens per task by type · Lower is better

  • Uncached input
  • Cached input
  • Output (reasoning included)
OpenEarthAgent Sentinel-2: tokens per taskNemotron 3 Ultra (high), Terminus-2: 19k; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 20k; Mistral Large 4 (high), Terminus-2: 25k; Claude Sonnet 5.5 (medium), Claude Code: 33k; Claude Sonnet 5.5 (low), Claude Code: 33k; Claude Opus 5.5 (low), Claude Code: 33k; Claude Sonnet 5.5 (high), Claude Code: 33k; Claude Opus 5.5 (medium), Claude Code: 35k; Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2: 36k; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 40k; Step 5 Preview (Reasoning), Terminus-2: 41k; Claude Opus 4.8 (high), Claude Code: 43k; Claude Opus 5.5 (high), Claude Code: 47k; Claude Sonnet 5.5 (xhigh), Claude Code: 50k; Claude Opus 5.5 (xhigh), Claude Code: 61k; Claude Opus 5 (high), Claude Code: 75k; Claude Haiku 4.5 (Reasoning), Claude Code: 84k; Claude Opus 4.7 (high), Claude Code: 93k; Claude Haiku 5.5 (low), Claude Code: 94k; Claude Sonnet 5 (high), Claude Code: 109k; Claude Haiku 5.5 (medium), Claude Code: 110k; Claude Haiku 5.5 (high), Claude Code: 135k; Claude Haiku 5.5 (xhigh), Claude Code: 162k; Claude Haiku 5.5 (max), Claude Code: 296k; Claude Opus 5.5 (max), Claude Code: 312k0100k200k300k400kNemotron 3 Ultra (high)Terminus-219kNemotron 3.5 Super VL 120B (Reasoning)Terminus-220kMistral Large 4 (high)Terminus-225kClaude Sonnet 5.5 (medium)Claude Code33kClaude Sonnet 5.5 (low)Claude Code33kClaude Opus 5.5 (low)Claude Code33kClaude Sonnet 5.5 (high)Claude Code33kClaude Opus 5.5 (medium)Claude Code35kNemotron 3 Nano Omni 30B (Reasoning)Terminus-236kNemotron 3.5 Lightning 30B (Reasoning)Terminus-240kStep 5 Preview (Reasoning)Terminus-241kClaude Opus 4.8 (high)Claude Code43kClaude Opus 5.5 (high)Claude Code47kClaude Sonnet 5.5 (xhigh)Claude Code50kClaude Opus 5.5 (xhigh)Claude Code61kClaude Opus 5 (high)Claude Code75kClaude Haiku 4.5 (Reasoning)Claude Code84kClaude Opus 4.7 (high)Claude Code93kClaude Haiku 5.5 (low)Claude Code94kClaude Sonnet 5 (high)Claude Code109kClaude Haiku 5.5 (medium)Claude Code110kClaude Haiku 5.5 (high)Claude Code135kClaude Haiku 5.5 (xhigh)Claude Code162kClaude Haiku 5.5 (max)Claude Code296kClaude Opus 5.5 (max)Claude Code312k
How to read this chart

What the parts mean. Uncached input: input tokens read at the full price. Cached input: input tokens served from the provider's cache at the lower cached price. Output: the tokens the model wrote, its reasoning included, as reported by the provider.

Compare with care. Models use different tokenisers, so the same text is a different number of tokens for different models. Use token counts to see how much work each model did, and cost to compare spending.

Data table: OpenEarthAgent Sentinel-2: tokens per task
OpenEarthAgent Sentinel-2: tokens per task
ModelDeveloperUncached inputCached inputOutput (reasoning included)Total
Nemotron 3 Ultra (high), Terminus-2NVIDIA15k2k2k19k
Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2NVIDIA14k3k3k20k
Mistral Large 4 (high), Terminus-2Mistral AI6k15k4k25k
Claude Sonnet 5.5 (medium), Claude CodeAnthropic5k28k68033k
Claude Sonnet 5.5 (low), Claude CodeAnthropic5k28k68733k
Claude Opus 5.5 (low), Claude CodeAnthropic5k27k68733k
Claude Sonnet 5.5 (high), Claude CodeAnthropic5k28k74233k
Claude Opus 5.5 (medium), Claude CodeAnthropic5k29k87135k
Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2NVIDIA8k22k5k36k
Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2NVIDIA17k18k4k40k
Step 5 Preview (Reasoning), Terminus-2StepFun3k31k6k41k
Claude Opus 4.8 (high), Claude CodeAnthropic5k37k95943k
Claude Opus 5.5 (high), Claude CodeAnthropic6k40k1k47k
Claude Sonnet 5.5 (xhigh), Claude CodeAnthropic6k43k1k50k
Claude Opus 5.5 (xhigh), Claude CodeAnthropic7k53k2k61k
Claude Opus 5 (high), Claude CodeAnthropic6k67k2k75k
Claude Haiku 4.5 (Reasoning), Claude CodeAnthropic6k75k3k84k
Claude Opus 4.7 (high), Claude CodeAnthropic6k86k1k93k
Claude Haiku 5.5 (low), Claude CodeAnthropic5k87k1k94k
Claude Sonnet 5 (high), Claude CodeAnthropic6k102k1k109k
Claude Haiku 5.5 (medium), Claude CodeAnthropic5k104k2k110k
Claude Haiku 5.5 (high), Claude CodeAnthropic7k126k2k135k
Claude Haiku 5.5 (xhigh), Claude CodeAnthropic8k150k3k162k
Claude Haiku 5.5 (max), Claude CodeAnthropic22k258k16k296k
Claude Opus 5.5 (max), Claude CodeAnthropic22k277k13k312k

Open weights

OpenEarthAgent Sentinel-2: tokens per task

Average tokens per task by type · Lower is better

  • Uncached input
  • Cached input
  • Output (reasoning included)
OpenEarthAgent Sentinel-2: tokens per taskNemotron 3 Ultra (high), Terminus-2: 19k; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 20k; Mistral Large 4 (high), Terminus-2: 25k; Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2: 36k; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 40k; Step 5 Preview (Reasoning), Terminus-2: 41k015k30k45k60kNemotron 3 Ultra (high)Terminus-219kNemotron 3.5 Super VL 120B (Reasoning)Terminus-220kMistral Large 4 (high)Terminus-225kNemotron 3 Nano Omni 30B (Reasoning)Terminus-236kNemotron 3.5 Lightning 30B (Reasoning)Terminus-240kStep 5 Preview (Reasoning)Terminus-241k
How to read this chart

What the parts mean. Uncached input: input tokens read at the full price. Cached input: input tokens served from the provider's cache at the lower cached price. Output: the tokens the model wrote, its reasoning included, as reported by the provider.

Compare with care. Models use different tokenisers, so the same text is a different number of tokens for different models. Use token counts to see how much work each model did, and cost to compare spending.

Data table: OpenEarthAgent Sentinel-2: tokens per task
OpenEarthAgent Sentinel-2: tokens per task
ModelDeveloperUncached inputCached inputOutput (reasoning included)Total
Nemotron 3 Ultra (high), Terminus-2NVIDIA15k2k2k19k
Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2NVIDIA14k3k3k20k
Mistral Large 4 (high), Terminus-2Mistral AI6k15k4k25k
Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2NVIDIA8k22k5k36k
Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2NVIDIA17k18k4k40k
Step 5 Preview (Reasoning), Terminus-2StepFun3k31k6k41k

Cost

All

OpenEarthAgent Sentinel-2: cost per task

Average cost per task (USD) by token type · Lower is better

  • Uncached input
  • Cached input
  • Output
  • Not split by token type
OpenEarthAgent Sentinel-2: cost per taskValues in US dollars: Claude Haiku 5.5 (low), Claude Code: $0.0025; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: $0.0027; Claude Haiku 5.5 (medium), Claude Code: $0.0028; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: $0.0036; Claude Haiku 5.5 (high), Claude Code: $0.0036; Claude Haiku 5.5 (xhigh), Claude Code: $0.0046; Mistral Large 4 (high), Terminus-2: $0.013; Nemotron 3 Ultra (high), Terminus-2: $0.015; Claude Haiku 5.5 (max), Claude Code: $0.021; Step 5 Preview (Reasoning), Terminus-2: $0.025; Claude Sonnet 5.5 (medium), Claude Code: $0.028; Claude Sonnet 5.5 (low), Claude Code: $0.029; Claude Sonnet 5.5 (high), Claude Code: $0.030; Claude Haiku 4.5 (Reasoning), Claude Code: $0.033; Claude Sonnet 5.5 (xhigh), Claude Code: $0.040; Claude Sonnet 5 (high), Claude Code: $0.058; Claude Opus 5.5 (low), Claude Code: $0.058; Claude Opus 5.5 (medium), Claude Code: $0.063; Claude Opus 5.5 (high), Claude Code: $0.075; Claude Opus 4.8 (high), Claude Code: $0.093; Claude Opus 5.5 (xhigh), Claude Code: $0.096; Claude Opus 4.7 (high), Claude Code: $0.13; Claude Opus 5 (high), Claude Code: $0.14; Claude Opus 5.5 (max), Claude Code: $0.49$0$0.15$0.30$0.45$0.60Claude Haiku 5.5 (low)Claude Code$0.0025Nemotron 3.5 Lightning 30B (Reasoning)Terminus-2$0.0027Claude Haiku 5.5 (medium)Claude Code$0.0028Nemotron 3.5 Super VL 120B (Reasoning)Terminus-2$0.0036Claude Haiku 5.5 (high)Claude Code$0.0036Claude Haiku 5.5 (xhigh)Claude Code$0.0046Mistral Large 4 (high)Terminus-2$0.013Nemotron 3 Ultra (high)Terminus-2$0.015Claude Haiku 5.5 (max)Claude Code$0.021Step 5 Preview (Reasoning)Terminus-2$0.025Claude Sonnet 5.5 (medium)Claude Code$0.028Claude Sonnet 5.5 (low)Claude Code$0.029Claude Sonnet 5.5 (high)Claude Code$0.030Claude Haiku 4.5 (Reasoning)Claude Code$0.033Claude Sonnet 5.5 (xhigh)Claude Code$0.040Claude Sonnet 5 (high)Claude Code$0.058Claude Opus 5.5 (low)Claude Code$0.058Claude Opus 5.5 (medium)Claude Code$0.063Claude Opus 5.5 (high)Claude Code$0.075Claude Opus 4.8 (high)Claude Code$0.093Claude Opus 5.5 (xhigh)Claude Code$0.096Claude Opus 4.7 (high)Claude Code$0.13Claude Opus 5 (high)Claude Code$0.14Claude Opus 5.5 (max)Claude Code$0.49
How to read this chart

What cost is measuring. What one attempt at a task would cost if paid token by token at the provider's published API prices, averaged over attempts: ordinary input tokens, input tokens served from the provider's cache at the lower cached price, tokens written to the cache where the provider charges for that, and output tokens (thinking included). Cost assumes each harness uses the model's pay-per-token API, priced at public list rates; sandboxes and our own engineering time are not included. Failed attempts count too.

How the cost is split. Each part is the entry's average tokens of that type times its list price. The rest, mostly tokens written to the cache, is shown as not split by token type.

Data table: OpenEarthAgent Sentinel-2: cost per task
OpenEarthAgent Sentinel-2: cost per task
ModelDeveloperUncached inputCached inputOutputNot split by token typeTotal
Claude Haiku 5.5 (low), Claude CodeAnthropic$0.00054$0.00087$0.00058$0.00054$0.0025
Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2NVIDIA$0.0010$0.00080$0.00089$0.00$0.0027
Claude Haiku 5.5 (medium), Claude CodeAnthropic$0.00049$0.0010$0.00077$0.00049$0.0028
Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2NVIDIA$0.0021$0.000062$0.0015$0.00$0.0036
Claude Haiku 5.5 (high), Claude CodeAnthropic$0.00067$0.0013$0.00099$0.00067$0.0036
Claude Haiku 5.5 (xhigh), Claude CodeAnthropic$0.00083$0.0015$0.0015$0.00083$0.0046
Mistral Large 4 (high), Terminus-2Mistral AI$0.0042$0.0011$0.0075$0.000000076$0.013
Nemotron 3 Ultra (high), Terminus-2NVIDIA$0.0092$0.00021$0.0055$0.00$0.015
Claude Haiku 5.5 (max), Claude CodeAnthropic$0.0022$0.0026$0.0080$0.0078$0.021
Step 5 Preview (Reasoning), Terminus-2StepFun$0.0037$0.0033$0.018$0.00000018$0.025
Claude Sonnet 5.5 (medium), Claude CodeAnthropic$0.0095$0.0028$0.0068$0.0095$0.028
Claude Sonnet 5.5 (low), Claude CodeAnthropic$0.0095$0.0028$0.0069$0.0095$0.029
Claude Sonnet 5.5 (high), Claude CodeAnthropic$0.0097$0.0028$0.0074$0.0097$0.030
Claude Haiku 4.5 (Reasoning), Claude CodeAnthropic$0.0063$0.0075$0.013$0.0063$0.033
Claude Sonnet 5.5 (xhigh), Claude CodeAnthropic$0.012$0.0043$0.012$0.012$0.040
Claude Sonnet 5 (high), Claude CodeAnthropic$0.012$0.020$0.013$0.012$0.058
Claude Opus 5.5 (low), Claude CodeAnthropic$0.020$0.0055$0.014$0.020$0.058
Claude Opus 5.5 (medium), Claude CodeAnthropic$0.020$0.0058$0.017$0.020$0.063
Claude Opus 5.5 (high), Claude CodeAnthropic$0.023$0.0080$0.022$0.023$0.075
Claude Opus 4.8 (high), Claude CodeAnthropic$0.025$0.019$0.024$0.025$0.093
Claude Opus 5.5 (xhigh), Claude CodeAnthropic$0.027$0.011$0.032$0.027$0.096
Claude Opus 4.7 (high), Claude CodeAnthropic$0.029$0.043$0.026$0.029$0.13
Claude Opus 5 (high), Claude CodeAnthropic$0.032$0.033$0.044$0.032$0.14
Claude Opus 5.5 (max), Claude CodeAnthropic$0.087$0.055$0.26$0.087$0.49
Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2NVIDIANo dataNo dataNo dataNo dataNo data

Open weights

OpenEarthAgent Sentinel-2: cost per task

Average cost per task (USD) by token type · Lower is better

  • Uncached input
  • Cached input
  • Output
  • Not split by token type
OpenEarthAgent Sentinel-2: cost per taskValues in US dollars: Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: $0.0027; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: $0.0036; Mistral Large 4 (high), Terminus-2: $0.013; Nemotron 3 Ultra (high), Terminus-2: $0.015; Step 5 Preview (Reasoning), Terminus-2: $0.025$0$0.01$0.02$0.03$0.04Nemotron 3.5 Lightning 30B (Reasoning)Terminus-2$0.0027Nemotron 3.5 Super VL 120B (Reasoning)Terminus-2$0.0036Mistral Large 4 (high)Terminus-2$0.013Nemotron 3 Ultra (high)Terminus-2$0.015Step 5 Preview (Reasoning)Terminus-2$0.025
How to read this chart

What cost is measuring. What one attempt at a task would cost if paid token by token at the provider's published API prices, averaged over attempts: ordinary input tokens, input tokens served from the provider's cache at the lower cached price, tokens written to the cache where the provider charges for that, and output tokens (thinking included). Cost assumes each harness uses the model's pay-per-token API, priced at public list rates; sandboxes and our own engineering time are not included. Failed attempts count too.

How the cost is split. Each part is the entry's average tokens of that type times its list price. The rest, mostly tokens written to the cache, is shown as not split by token type.

Data table: OpenEarthAgent Sentinel-2: cost per task
OpenEarthAgent Sentinel-2: cost per task
ModelDeveloperUncached inputCached inputOutputNot split by token typeTotal
Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2NVIDIA$0.0010$0.00080$0.00089$0.00$0.0027
Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2NVIDIA$0.0021$0.000062$0.0015$0.00$0.0036
Mistral Large 4 (high), Terminus-2Mistral AI$0.0042$0.0011$0.0075$0.000000076$0.013
Nemotron 3 Ultra (high), Terminus-2NVIDIA$0.0092$0.00021$0.0055$0.00$0.015
Step 5 Preview (Reasoning), Terminus-2StepFun$0.0037$0.0033$0.018$0.00000018$0.025
Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2NVIDIANo dataNo dataNo dataNo dataNo data

Speed

All

OpenEarthAgent Sentinel-2: time per task

Time per task (at 100 output tokens/s), minutes · Lower is better

OpenEarthAgent Sentinel-2: time per taskOpenEarthAgent Sentinel-2: time per task: Claude Opus 5.5 (low), Claude Code: 0.2 min; Claude Sonnet 5.5 (low), Claude Code: 0.2 min; Claude Sonnet 5.5 (high), Claude Code: 0.2 min; Claude Sonnet 5.5 (medium), Claude Code: 0.2 min; Claude Opus 5.5 (medium), Claude Code: 0.2 min; Claude Opus 4.8 (high), Claude Code: 0.3 min; Claude Opus 4.7 (high), Claude Code: 0.3 min; Claude Opus 5.5 (high), Claude Code: 0.3 min; Claude Haiku 5.5 (low), Claude Code: 0.3 min; Claude Sonnet 5 (high), Claude Code: 0.3 min; Claude Sonnet 5.5 (xhigh), Claude Code: 0.3 min; Claude Haiku 5.5 (medium), Claude Code: 0.4 min; Claude Opus 5.5 (xhigh), Claude Code: 0.4 min; Claude Opus 5 (high), Claude Code: 0.4 min; Claude Haiku 5.5 (high), Claude Code: 0.4 min; Nemotron 3 Ultra (high), Terminus-2: 0.5 min; Claude Haiku 4.5 (Reasoning), Claude Code: 0.5 min; Claude Haiku 5.5 (xhigh), Claude Code: 0.6 min; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 1.0 min; Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2: 1.0 min; Step 5 Preview (Reasoning), Terminus-2: 1.3 min; Mistral Large 4 (high), Terminus-2: 1.6 min; Claude Opus 5.5 (max), Claude Code: 2.3 min; Claude Haiku 5.5 (max), Claude Code: 2.8 min; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 10.4 min0 min5 min10 min15 min20 minClaude Opus 5.5 (low)Claude Code0.2 minClaude Sonnet 5.5 (low)Claude Code0.2 minClaude Sonnet 5.5 (high)Claude Code0.2 minClaude Sonnet 5.5 (medium)Claude Code0.2 minClaude Opus 5.5 (medium)Claude Code0.2 minClaude Opus 4.8 (high)Claude Code0.3 minClaude Opus 4.7 (high)Claude Code0.3 minClaude Opus 5.5 (high)Claude Code0.3 minClaude Haiku 5.5 (low)Claude Code0.3 minClaude Sonnet 5 (high)Claude Code0.3 minClaude Sonnet 5.5 (xhigh)Claude Code0.3 minClaude Haiku 5.5 (medium)Claude Code0.4 minClaude Opus 5.5 (xhigh)Claude Code0.4 minClaude Opus 5 (high)Claude Code0.4 minClaude Haiku 5.5 (high)Claude Code0.4 minNemotron 3 Ultra (high)Terminus-20.5 minClaude Haiku 4.5 (Reasoning)Claude Code0.5 minClaude Haiku 5.5 (xhigh)Claude Code0.6 minNemotron 3.5 Lightning 30B (Reasoning)Terminus-21.0 minNemotron 3 Nano Omni 30B (Reasoning)Terminus-21.0 minStep 5 Preview (Reasoning)Terminus-21.3 minMistral Large 4 (high)Terminus-21.6 minClaude Opus 5.5 (max)Claude Code2.3 minClaude Haiku 5.5 (max)Claude Code2.8 minNemotron 3.5 Super VL 120B (Reasoning)Terminus-210.4 min
How to read this chart

Time per task. Time as if every model wrote 100 output tokens per second: output tokens divided by 100, plus the time the agent's commands took. Serving speed differs between providers and is not ranked.

Data table: OpenEarthAgent Sentinel-2: time per task

Open weights

OpenEarthAgent Sentinel-2: time per task

Time per task (at 100 output tokens/s), minutes · Lower is better

OpenEarthAgent Sentinel-2: time per taskOpenEarthAgent Sentinel-2: time per task: Nemotron 3 Ultra (high), Terminus-2: 0.5 min; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 1.0 min; Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2: 1.0 min; Step 5 Preview (Reasoning), Terminus-2: 1.3 min; Mistral Large 4 (high), Terminus-2: 1.6 min; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 10.4 min0 min5 min10 min15 min20 minNemotron 3 Ultra (high)Terminus-20.5 minNemotron 3.5 Lightning 30B (Reasoning)Terminus-21.0 minNemotron 3 Nano Omni 30B (Reasoning)Terminus-21.0 minStep 5 Preview (Reasoning)Terminus-21.3 minMistral Large 4 (high)Terminus-21.6 minNemotron 3.5 Super VL 120B (Reasoning)Terminus-210.4 min
How to read this chart

Time per task. Time as if every model wrote 100 output tokens per second: output tokens divided by 100, plus the time the agent's commands took. Serving speed differs between providers and is not ranked.

Data table: OpenEarthAgent Sentinel-2: time per task

All

OpenEarthAgent Sentinel-2: score vs. time per task

OpenEarthAgent Sentinel-2 score against time per task (at 100 output tokens/s), minutes · Higher and further left is better

  • Most attractive quadrant
  • Pareto line
OpenEarthAgent Sentinel-2: score vs. time per taskOpenEarthAgent Sentinel-2 score (0 to 100) against time per task (minutes), log scale: Mistral Large 4 (high), Terminus-2: 99.18, 1.6 min; Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2: 52.46, 1.0 min; Nemotron 3 Ultra (high), Terminus-2: 100, 0.5 min; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 93.44, 1.0 min; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 96.72, 10.4 min; Step 5 Preview (Reasoning), Terminus-2: 100, 1.3 min; Claude Haiku 4.5 (Reasoning), Claude Code: 100, 0.5 min; Claude Opus 5.5 (medium), Claude Code: 100, 0.2 min; Claude Opus 5.5 (high), Claude Code: 100, 0.3 min; Claude Sonnet 5.5 (medium), Claude Code: 100, 0.2 min; Claude Sonnet 5.5 (high), Claude Code: 100, 0.2 min; Claude Haiku 5.5 (medium), Claude Code: 100, 0.4 min; Claude Haiku 5.5 (high), Claude Code: 100, 0.4 min; Claude Opus 5.5 (xhigh), Claude Code: 100, 0.4 min; Claude Sonnet 5.5 (xhigh), Claude Code: 100, 0.3 min; Claude Haiku 5.5 (xhigh), Claude Code: 100, 0.6 min; Claude Opus 5.5 (low), Claude Code: 100, 0.2 min; Claude Sonnet 5.5 (low), Claude Code: 100, 0.2 min; Claude Haiku 5.5 (low), Claude Code: 100, 0.3 min; Claude Opus 5.5 (max), Claude Code: 100, 2.3 min; Claude Haiku 5.5 (max), Claude Code: 100, 2.8 min; Claude Opus 5 (high), Claude Code: 100, 0.4 min; Claude Sonnet 5 (high), Claude Code: 100, 0.3 min; Claude Opus 4.8 (high), Claude Code: 100, 0.3 min; Claude Opus 4.7 (high), Claude Code: 100, 0.3 minMost attractive quadrant4050607080901000.1 min0.2 min0.5 min1 min2 min5 min10 min20 minTime per task (minutes, log scale)OpenEarthAgent Sentinel-2 scoreNemotron 3 Ultra (high), Terminus-2Step 5 Preview (Reasoning), Terminus-2Claude Haiku 4.5 (Reasoning), Claude CodeClaude Opus 5.5 (low), Claude CodeClaude Opus 5.5 (medium), Claude CodeClaude Opus 5.5 (high), Claude CodeClaude Sonnet 5.5 (medium), Claude CodeClaude Sonnet 5.5 (high), Claude CodeClaude Haiku 5.5 (medium), Claude CodeClaude Haiku 5.5 (high), Claude CodeClaude Opus 5.5 (xhigh), Claude CodeClaude Sonnet 5.5 (xhigh), Claude CodeClaude Haiku 5.5 (xhigh), Claude CodeClaude Sonnet 5.5 (low), Claude CodeClaude Haiku 5.5 (low), Claude CodeClaude Opus 5.5 (max), Claude CodeClaude Haiku 5.5 (max), Claude CodeClaude Opus 5 (high), Claude CodeClaude Sonnet 5 (high), Claude CodeClaude Opus 4.8 (high), Claude CodeClaude Opus 4.7 (high), Claude CodeMistral Large 4 (high), Terminus-2Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2
How to read this chart

Time per task. Time as if every model wrote 100 output tokens per second: output tokens divided by 100, plus the time the agent's commands took. Serving speed differs between providers and is not ranked.

How to read this chart. Each point is one entry. Further left means a lower average time per task (log scale); higher means a higher OpenEarthAgent Sentinel-2 score. Both axes fit the points shown, so the score axis does not start at zero. The shaded most attractive quadrant is the top-left quarter of the plot: a reading aid, not a fixed bar. The dotted Pareto line joins the points that no faster one beats on score.

Data table: OpenEarthAgent Sentinel-2: score vs. time per task

Open weights

OpenEarthAgent Sentinel-2: score vs. time per task

OpenEarthAgent Sentinel-2 score against time per task (at 100 output tokens/s), minutes · Higher and further left is better

  • Most attractive quadrant
  • Pareto line
OpenEarthAgent Sentinel-2: score vs. time per taskOpenEarthAgent Sentinel-2 score (0 to 100) against time per task (minutes), log scale: Mistral Large 4 (high), Terminus-2: 99.18, 1.6 min; Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2: 52.46, 1.0 min; Nemotron 3 Ultra (high), Terminus-2: 100, 0.5 min; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 93.44, 1.0 min; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 96.72, 10.4 min; Step 5 Preview (Reasoning), Terminus-2: 100, 1.3 minMost attractive quadrant4050607080901000.2 min0.5 min1 min2 min5 min10 min20 minTime per task (minutes, log scale)OpenEarthAgent Sentinel-2 scoreNemotron 3 Ultra (high), Terminus-2Step 5 Preview (Reasoning), Terminus-2Mistral Large 4 (high), Terminus-2Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2
How to read this chart

Time per task. Time as if every model wrote 100 output tokens per second: output tokens divided by 100, plus the time the agent's commands took. Serving speed differs between providers and is not ranked.

How to read this chart. Each point is one entry. Further left means a lower average time per task (log scale); higher means a higher OpenEarthAgent Sentinel-2 score. Both axes fit the points shown, so the score axis does not start at zero. The shaded most attractive quadrant is the top-left quarter of the plot: a reading aid, not a fixed bar. The dotted Pareto line joins the points that no faster one beats on score.

Data table: OpenEarthAgent Sentinel-2: score vs. time per task
OpenEarthAgent Sentinel-2: score vs. time per task
ModelOpenEarthAgent Sentinel-2 scoreTime per task (minutes)On the Pareto line
Nemotron 3 Ultra (high), Terminus-21000.5 minYes
Step 5 Preview (Reasoning), Terminus-21001.3 minNo
Mistral Large 4 (high), Terminus-2991.6 minNo
Nemotron 3.5 Super VL 120B (Reasoning), Terminus-29710.4 minNo
Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2931.0 minNo
Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2521.0 minNo

Score vs. release date

All

OpenEarthAgent Sentinel-2: score vs. release date

OpenEarthAgent Sentinel-2 score against the model's public release date

OpenEarthAgent Sentinel-2: score vs. release dateOpenEarthAgent Sentinel-2 score (0 to 100) against release date: Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2: 52.46, 28 April 2026; Nemotron 3 Ultra (high), Terminus-2: 100, 4 June 2026; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 93.44, 11 August 2026; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 96.72, 15 October 2026; Claude Haiku 4.5 (Reasoning), Claude Code: 100, 15 October 2025; Claude Opus 5.5 (medium), Claude Code: 100, 22 September 2026; Claude Opus 5.5 (high), Claude Code: 100, 22 September 2026; Claude Sonnet 5.5 (medium), Claude Code: 100, 28 September 2026; Claude Sonnet 5.5 (high), Claude Code: 100, 28 September 2026; Claude Haiku 5.5 (medium), Claude Code: 100, 7 October 2026; Claude Haiku 5.5 (high), Claude Code: 100, 7 October 2026; Claude Opus 5.5 (xhigh), Claude Code: 100, 22 September 2026; Claude Sonnet 5.5 (xhigh), Claude Code: 100, 28 September 2026; Claude Haiku 5.5 (xhigh), Claude Code: 100, 7 October 2026; Claude Opus 5.5 (low), Claude Code: 100, 22 September 2026; Claude Sonnet 5.5 (low), Claude Code: 100, 28 September 2026; Claude Haiku 5.5 (low), Claude Code: 100, 7 October 2026; Claude Opus 5.5 (max), Claude Code: 100, 22 September 2026; Claude Haiku 5.5 (max), Claude Code: 100, 7 October 2026; Claude Opus 5 (high), Claude Code: 100, 24 July 2026; Claude Sonnet 5 (high), Claude Code: 100, 30 June 2026; Claude Opus 4.8 (high), Claude Code: 100, 28 May 2026; Claude Opus 4.7 (high), Claude Code: 100, 16 April 2026405060708090100Oct 2025Jan 2026Apr 2026Jul 2026Oct 2026Jan 2027Release dateOpenEarthAgent Sentinel-2 scoreNemotron 3 Ultra (high), Terminus-2Claude Haiku 4.5 (Reasoning), Claude CodeClaude Opus 5.5 (medium), Claude CodeClaude Opus 5.5 (high), Claude CodeClaude Sonnet 5.5 (medium), Claude CodeClaude Sonnet 5.5 (high), Claude CodeClaude Haiku 5.5 (medium), Claude CodeClaude Haiku 5.5 (high), Claude CodeClaude Opus 5.5 (xhigh), Claude CodeClaude Sonnet 5.5 (xhigh), Claude CodeClaude Haiku 5.5 (xhigh), Claude CodeClaude Opus 5.5 (low), Claude CodeClaude Sonnet 5.5 (low), Claude CodeClaude Haiku 5.5 (low), Claude CodeClaude Opus 5.5 (max), Claude CodeClaude Haiku 5.5 (max), Claude CodeClaude Opus 5 (high), Claude CodeClaude Sonnet 5 (high), Claude CodeClaude Opus 4.8 (high), Claude CodeClaude Opus 4.7 (high), Claude CodeNemotron 3.5 Super VL 120B (Reasoning), Terminus-2Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2
How to read this chart

How to read this chart. Each point is one entry: across, the date its model was first released; up, its OpenEarthAgent Sentinel-2 score. Models with no published release date are not plotted.

Data table: OpenEarthAgent Sentinel-2: score vs. release date

Open weights

OpenEarthAgent Sentinel-2: score vs. release date

OpenEarthAgent Sentinel-2 score against the model's public release date

OpenEarthAgent Sentinel-2: score vs. release dateOpenEarthAgent Sentinel-2 score (0 to 100) against release date: Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2: 52.46, 28 April 2026; Nemotron 3 Ultra (high), Terminus-2: 100, 4 June 2026; Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2: 93.44, 11 August 2026; Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2: 96.72, 15 October 2026405060708090100Mar 2026May 2026Jul 2026Sep 2026Nov 2026Release dateOpenEarthAgent Sentinel-2 scoreNemotron 3 Ultra (high), Terminus-2Nemotron 3.5 Super VL 120B (Reasoning), Terminus-2Nemotron 3.5 Lightning 30B (Reasoning), Terminus-2Nemotron 3 Nano Omni 30B (Reasoning), Terminus-2
How to read this chart

How to read this chart. Each point is one entry: across, the date its model was first released; up, its OpenEarthAgent Sentinel-2 score. Models with no published release date are not plotted.

Data table: OpenEarthAgent Sentinel-2: score vs. release date
OpenEarthAgent Sentinel-2: score vs. release date
ModelOpenEarthAgent Sentinel-2 scoreRelease date
Nemotron 3 Ultra (high), Terminus-21004 June 2026
Step 5 Preview (Reasoning), Terminus-2100No data
Mistral Large 4 (high), Terminus-299No data
Nemotron 3.5 Super VL 120B (Reasoning), Terminus-29715 October 2026
Nemotron 3.5 Lightning 30B (Reasoning), Terminus-29311 August 2026
Nemotron 3 Nano Omni 30B (Reasoning), Terminus-25228 April 2026

Background

OpenEarthAgent Sentinel-2 asks how a place changed between two dates: vegetation loss and gain, urban growth, or burn severity. The agent gets the two Sentinel-2 index rasters the source's tools made for that place and the source's class breaks, computes the change itself and reports the share of the area in each class.

Licence: dataset Apache 2.0 as its card declares; contains modified Copernicus Sentinel-2 data.

Published by MBZUAI.

Related links

Explore evaluations