Short coding task
Build a CSV parser
Flavio Copes sent both models the same prompt through OpenRouter at high effort. Both passed all 27 tests.
- Neon
- $0.091 49.3 s
- Opus 5.5
- $0.029 12.4 s
Opus used fewer output tokens on this run. Test and results
Try Neon for long inputs and short answers. For small coding changes, test Opus 5.5 first. Compare both on your own workload before switching a multi-step agent.
Standard rates, context, and published results by thinking level.
| Metric | Gemini 4 Neon | Claude Opus 5.5 |
|---|---|---|
| Input price per 1M tokens | $3.00 | $4.00 |
| Output price per 1M tokens | $15.00 | $20.00 |
| Context window | 1M tokens | 1M tokens |
| Inputs | Text, images, PDFs, audio, video | Text, images, PDFs |
| Default effort | High | Medium |
| AA Intelligence Index published effort levels | 44 low 55 med 60 high 61 xhigh | 35 low 59 max |
| Terminal-Bench 4.1 cheating-adjusted accuracy | 65.2% | 65.8% |
Benchmark snapshot from November 2026. The reported Terminal-Bench scores come from different agent setups. Artificial Analysis · Evaluation methods
Enter the same billed usage for both models. The estimates update as you type.
Enter values in millions of tokens. For example, 0.2 = 200,000.
/ mo
/ mo
Neon costs $20.00 less per month at this usage.
USD, standard on-demand API rates. Include billed thinking tokens in output usage. Estimates exclude cache and batch discounts, tool fees, surcharges, and taxes. Google pricing · Anthropic pricing
In independent tests, the cost ranking changed when the task changed.
Short coding task
Flavio Copes sent both models the same prompt through OpenRouter at high effort. Both passed all 27 tests.
Opus used fewer output tokens on this run. Test and results
Long-context retrieval
Both models returned the two correct articles from the same set of 23 posts.
Neon counted about 103k input tokens, and Opus about 140k. Test and results
Community comparison
@notjazii used the same prompt at each model's highest effort. The author preferred Opus’s result. The reported run cost was $3.50 for Neon and $7.20 for Opus.
For long documents with short answers, try Gemini 4 Neon and measure retrieval accuracy. For small coding changes, give Opus 5.5 the same task and compare time, test passes, and billed tokens. Use your own repository for agent workflows.
Standard input and output rates per million tokens are $3.00 and $15.00 for Gemini 4 Neon, and $4.00 and $20.00 for Claude Opus 5.5. Enter your billed token counts to estimate an API bill.
Both support about one million context tokens. Output length, accepted inputs, and access to extended generation depend on the model and API mode.
Use the provider’s billed output-token count, including thinking or reasoning tokens charged as output. The visible answer often accounts for only part of the output bill.
It estimates standard on-demand API charges in USD. Cache reads and writes, batch and priority tiers, regional surcharges, tool charges, taxes, and app subscription limits are outside this estimate.
Benchmark versions, effort, and the tools available to the agent can all change results. Compare matched setups when choosing a model for your workflow.