Create more. Sell more this Black Friday.Explore
Skip to article
Gemini 4 Neon — Glowing cyan glass beside a graphite block
Comparison

Gemini 4 Neon vs Gemini 4 Carbon for Coding, Cost, and Routing

Compare Neon and Carbon on coding, migration, and automation. See how the December price change and retries affect task cost, then test a routing policy.

EvoLink TeamProduct Team
9 min read
Start with Gemini 4 Neon high for frequent, well-scoped engineering tasks. Keep Carbon as the baseline for difficult architecture work and workflows it already handles reliably. On CursorBench 4.1, Neon high scored 63.7% versus Carbon high at 61.5% and cost less per task in that benchmark. Use your account's Carbon rate and both models' success on your own failure cases to decide how to route traffic. CursorBench 4.1

Which tasks suit Gemini 4 Neon and Carbon?

JobStarting choiceWhat would change the decision
Recurring cross-file bug with a reproducible failureNeon highCarbon consistently resolves cases Neon leaves incomplete
Ambiguous architecture change with costly reworkYour validated model, with Carbon as a baselineNeon meets the same integration checks and review bar at lower total cost
Monthly analysis with fixed outputs and traceable sourcesEvaluate Neon alongside your current routeA cheaper configuration delivers equally reliable source references
A failed job with missing inputs or a broken toolResolve the missing input or tool failureRerun once the prerequisite is restored

An invoice rounding bug gives the agent a clear target. The failing test shows where the totals diverge, and the patch has to preserve historical invoices and refund behavior. Reworking how pricing and billing services share state calls for decisions about boundaries, recovery, and compatibility. Unit tests can cover only part of that work, so keep Carbon as a baseline for a change like that.

What changes at higher thinking levels?

CursorBench 4.1 uses ambiguous, multi-file tasks drawn from real Cursor sessions. Its configurations give teams several starting points.

CursorBench 4.1 configurationScoreListed cost per task
Neon medium57.7%$3.07
Carbon medium60.4%$7.22
Carbon high61.5%$8.52
Neon high63.7%$6.54
Neon xhigh65.4%$10.67
For these tasks, moving Neon from high to xhigh added 1.7 percentage points and $4.13 per task. Test xhigh when you can identify difficult jobs before they run, then compare it against Carbon with the same checks and time limit. Cursor's costs reflect its own benchmark setup. CursorBench results and methodology
CursorBench 4.1 shows score versus listed task cost. Neon high scored 63.7% at $6.54 per task and Carbon high scored 61.5% at $8.52.
CursorBench 4.1 shows score versus listed task cost. Neon high scored 63.7% at $6.54 per task and Carbon high scored 61.5% at $8.52.
In the launch-week Artificial Analysis Intelligence Index snapshot, Neon high and Carbon high both rounded to 60. Neon took 14.8 minutes and $3.74 per task on average. Carbon averaged 37.1 minutes, with a cost of $7.76 at regular rates. In Plotly's separate data analysis benchmark, Neon was faster and cost more. Record latency and spend separately. Figures checked on November 19, 2026. Artificial Analysis model data · Plotly data-analysis benchmark

Why benchmarks disagree

CursorBench measures success on multi-file work in an editor. For a different workload, find a benchmark whose output resembles your own. The results below use different tasks and configurations, so compare the two models within each row.
EvaluationNeonCarbonDecision it informs
Vals Code Migration v1.1, Vals Index subset75.8%80.1%Include Carbon when a language migration must preserve behavior
Zapier AutomationBench, both high54.3%, $0.88/task estimated53.9%, $1.94/task estimatedCheck whether Neon completes routine tool workflows at lower cost
Plotly data analysis43/43, 46 s median, $17.03/run42/43, 93 s median, $11.55/runBudget speed and cost separately
On Plotly's test, Neon used fewer tokens and model calls, yet cost more at the rates applied to that run. Plotly raised Neon's score after correcting a grading rule and checking the answers by hand. Version your fixtures and rubric so that a grading change is visible in the results. Zapier's costs are estimates from its metrics export and use an assumed cache-read price. Vals methodology and scores · Zapier report · Plotly report
In Plotly’s test, the median question took 46 seconds for Neon and 93 for Carbon. The full runs cost $17.03 and $11.55, respectively.
In Plotly’s test, the median question took 46 seconds for Neon and 93 for Carbon. The full runs cost $17.03 and $11.55, respectively.

How Carbon’s December price change affects your budget

These are Google's standard API prices in USD per million tokens.

Model and periodInputOutputCached input
Neon$3$15$0.15
Carbon through November 30, 2026$2$10$0.10
Carbon from December 1, 2026$4$20$0.20
At 100,000 uncached input tokens and 10,000 billable output tokens, the same usage costs $0.45 on Neon, $0.30 on promotional Carbon, and $0.60 on regular-price Carbon. For an actual decision, use measured token consumption, including thinking and retries. On EvoLink, run the calculation with your account rates for both models. Google Carbon pricing · Google Neon pricing · EvoLink Carbon rates
At Google’s standard API rates, the same token mix costs $0.45 on Neon, $0.30 on Carbon through November 30, and $0.60 on Carbon from December 1.
At Google’s standard API rates, the same token mix costs $0.45 on Neon, $0.30 on Carbon through November 30, and $0.60 on Carbon from December 1.

Where retries erase the cost advantage

At Google's standard API rates, assume uncached input, cached input, and billable output make up the same proportions for both models. Through November, Neon costs less only below two thirds of Carbon's billable volume. At Carbon's December rate, costs are equal when Neon uses one third more. For a real workload with a changing token mix, price each category separately. Neon rates · Carbon rates
For a rerouting example, suppose a Neon attempt costs $0.36 and the complete Carbon attempt costs $0.62. If 20% of requests also need Carbon, starting with Neon costs $0.484 per incoming task on average, before review. If about 42% need Carbon, that policy is already around the cost of starting with Carbon. A higher eventual acceptance rate could still justify it. Compare the final cost per accepted task with the same time and quality requirements.

Use failure reasons to decide which tasks go to Carbon. A patch that passes unit tests but changes an established refund total is a candidate for a second pass if the model has enough evidence to correct it. A missing database export, unavailable tool, or denied permission needs the prerequisite restored. A second model call in those cases generally adds spend without addressing the failure.

Migrate one class of tasks at a time

Take recent tasks your existing model has already run. Group them by deliverable. A small patch, a diagnosis across several files, and a design with compatibility constraints need different checks. Replay each task from the same repository snapshot with the same tools and acceptance checks.

Record first-pass acceptance, elapsed time, retries, and the total bill. If Neon saves time on routine cross-file fixes and preserves acceptance, move that class first. If Carbon resolves a failure Neon repeatedly misses, keep Carbon on that class. Once both routes are available to your account, an EvoLink integration gives you a common place to manage model selection while keeping your application’s test and review rules. EvoLink Gemini collection · Gemini integration guide
See the Neon release guide for access and model limits, or the Opus 5.5 comparison if the decision also involves Claude Code.

Use hard cases your baseline handles to set routing rules

Build a replay set that covers well-scoped code fixes, ambiguous failures across components, and long changes that must preserve public behavior. Include past near-misses, such as a patch that changed a default, an old fixture that hid a regression, or a plausible report based on a stale source. For each task, save the starting revision, visible instructions, allowed tools, and what qualifies as finished.

Record the median and slow-end completion time for each configuration, including tool waits and retries. If a request is time-sensitive, look at the long-running tail before routing live traffic. If work runs overnight, accepted-task cost and the morning review queue may matter more than median latency. Record uncached input, cache hits, billable output, and human review separately so you can price the same trace under a later rate schedule.

Compare direct calls with a single escalation

Keep one Carbon-only baseline. Compare it with Neon alone and with Neon followed by one Carbon attempt after a detected, recoverable failure. Keep the task budget, permissions, and acceptance tests fixed. If the two-model path is valuable, its logs tell you which task class to move and which failure triggers actually pay for themselves. If almost every attempt escalates, start that class on Carbon and note the evidence for the rule.

FAQ

Is Neon always cheaper than Carbon?

Google's per-token rates favor Carbon during its introductory period and Neon at regular rates. Task cost also depends on usage and retries.

Which is the better first test for multi-file coding?

Neon high is a practical candidate for recurring work with clear checks. Include your current Carbon configuration.

Do both have a one-million-token context?

Google documents a shared one-million-token window for both. Check output and continuation limits on your chosen interface.

Do they have the same default thinking level?

Google documents high for Neon and medium for Carbon.

Should I send hard cases to Neon xhigh or Carbon?

Compare the two on the failure modes you actually see and record whether the extra attempt resolves the work.

Should I switch all traffic at once?

Start with the task class where you can measure acceptance and roll back quickly. Google's Gemini 4 developer guide

Which model should start a language migration?

Keep Carbon in the evaluation. It scored 80.1% versus Neon's 75.8% on the Code Migration subset used by the Vals Index. Include your team's compatibility fixtures. Vals Index

Can Carbon return shorter segments than Neon?

Google's Long Decode documentation distinguishes the synchronous segments. Neon permits up to 65,536 tokens per hop, while Carbon's non-streaming hops can cap at 32,768. Check the API and streaming mode you intend to use. Google Long Decode

Connect AI models through one API

Compare models and current pricing, then follow the API documentation to integrate your chosen model.