Gemini 4 Neon vs Gemini 4 Carbon: Complete Comparison and Report on Pricing, Cost per Task, Thinking Levels, and Benchmarks

Gemini 4 Neon is the model Google DeepMind released on November 18, 2026, as the middle tier of the Gemini 4 family, placed between Gemini 4 Hydrogen for fast everyday execution and Gemini 4 Carbon, which Google keeps describing as its option for the hardest long-horizon work. Carbon arrived first: Google announced it on October 28 for a restricted group of cyber defenders and opened it to developers, businesses and Google AI Ultra subscribers on November 10. Neon's own model card states that it is adapted from Carbon, and both models share the same 1,048,576-token unified context window, the same March 2026 knowledge cutoff and the same request schema, so a developer can switch between them by changing only the model ID.
The verdict on the published evidence favors Neon for most paid workloads. On the Gemini API, Neon costs $3 per 1M input tokens and $15 per 1M output tokens, which is 25% below Carbon's standard $4/$20 rate that takes effect on December 1, 2026, and 50% above Carbon's introductory $2/$10 rate that runs through November 30. Artificial Analysis measured Neon at xhigh effort at 61 on its Intelligence Index against 60 for Carbon at high, with a cost per task of $6.44 against $7.76, a task time of 21.5 minutes against 37.2, and output speed of about 110 tokens per second against 55. Vals AI places Neon second of 42 models on the Vals Index at 70.87%, with Carbon fourth at 70.45%.
Carbon keeps a lead on a narrower set of measurements: Terminal-Bench 4.0 in both independent tests, CritPt v2, Agents' Last Exam v1.1 in Google's own figures, and long-context and computer-use results that Google has published only for Carbon. For teams that spend heavily before November 30, Carbon is also the cheaper model for the next eleven days. Neon is the better default for agent work, coding agents, finance and legal analysis, and anyone on Google AI Pro, because Pro subscribers get Neon in the Gemini app and Antigravity while Carbon in the consumer apps is tied to Google AI Ultra. All prices in this report were checked on Google's official pricing and developer pages on November 19, 2026.
··········
GEMINI 4 NEON AND GEMINI 4 CARBON RELEASE TIMELINE AND MODEL LINEAGE.
The two models were released three weeks apart and come from the same line of development. Google's model cards describe a chain: Gemini 4 Carbon "is based on Gemini 4 Argon", and Gemini 4 Neon "is adapted from Gemini 4 Carbon". Neither card discloses the architecture, the parameter count or whether the models use a mixture-of-experts design; both defer training data, hardware and software details to the earlier cards.
Carbon's path to general availability was unusual. Google introduced it on October 28 as its "strongest model with state-of-the-art cybersecurity and long horizon capabilities" and gave it first to a subset of cyber defenders through the Fairwind Program, with developers and consumers promised access later. That broader release came on November 10, when Carbon reached the paid Gemini API tier, Google AI Studio, Android Studio, Gemini for business, Antigravity and Google AI Ultra. Neon skipped the restricted phase entirely and launched on November 18 across the API, AI Studio, the Gemini app, Antigravity and Gemini for business on the same day, with Fairwind members receiving a cyber variant over the following days. The launch details are covered in our report on Google's Gemini 4 Neon launch at $3/$15.
The API model IDs are gemini-4-neon and gemini-4-carbon, neither model has open weights, and both accept text, image, audio and video input with text output. Neon is Google DeepMind's fourth model release in seven weeks, according to Artificial Analysis, after Argon, Carbon and Hydrogen. Google has not said that Neon replaces Carbon, and Carbon remains listed and priced on every official page checked for this report.
··········
GEMINI 4 NEON VS GEMINI 4 CARBON API PRICING BEFORE AND AFTER DECEMBER 1.
The price comparison changes on a fixed date. Carbon launched with an introductory rate equal to Argon's, $2 per 1M input tokens and $10 per 1M output tokens, and Google's developer page states that the rate rises to $4 and $20 on December 1, 2026. Neon has a single flat rate with no introductory period. Output prices for both models include thinking tokens, and both bill internal tool instructions and function declarations when grounding is used.
........
Price item (USD per 1M tokens) | Gemini 4 Neon | Gemini 4 Carbon to Nov 30, 2026 | Gemini 4 Carbon from Dec 1, 2026 |
Input | $3.00 | $2.00 | $4.00 |
Cached input (implicit caching) | $0.15 | $0.10 | $0.20 |
Output, including thinking | $15.00 | $10.00 | $20.00 |
Free tier | Not available | Not available | Not available |
Long-context surcharge | None (flat across 1M) | None listed | None listed |
Grounding with Google Search | 5,000 free requests a month shared across Gemini 3 and newer models, then $14 per 1,000 | Same | Same |
Google Cloud batch, global (input / output) | $1.50 / $7.50 | $1.00 / $5.00 | $2.00 / $10.00 |
Google Cloud priority, global (input / output) | $5.40 / $27.00 | $3.60 / $18.00 | $7.20 / $36.00 |
........
Google Cloud's Gemini Enterprise Agent Platform uses the same base rates on its global endpoints and adds tiers the Gemini API developer page for Neon does not list. Regional (non-global) endpoints carry a 10% premium: $3.30/$16.50 for Neon, and $2.20/$11.00 rising to $4.40/$22.00 for Carbon. Google marks Carbon's pre-December Cloud rates as promotional, and Carbon has no cache read or cache write price on the Cloud page, while Neon lists $0.15 for cache reads and $3.75 for cache writes effective January 12, 2027.
Two Cloud billing notes apply to both models. Google says cache writes on Gemini 4 models are billed at 1.25 times the standard input price, and since November 13, 2026, completions longer than 32K tokens are billed for retrievable tokens even when the request later fails. Our guide to Gemini pricing and subscription features covers the consumer plans and the rest of the Gemini API price list.
··········
CONTEXT WINDOW, OUTPUT LIMITS AND THINKING LEVELS IN GEMINI 4 NEON AND CARBON.
The two models share the same token budget design. Input, thinking and output draw on one window of 1,048,576 tokens, and max_output_tokens caps thinking and visible output together. A synchronous response stops at 65,536 tokens; longer outputs require background execution with max_output_tokens set above that figure, in which case a response can run to the full window minus the input. Google's Cloud documentation adds that Carbon's non-streaming foreground segment is capped at 32,768 tokens, while the Gemini API page for Neon states that its 65,536-token limit applies to both streaming and non-streaming requests.
The real difference is the thinking scale. Carbon offers three levels and defaults to medium; Neon offers four, defaults to high and adds xhigh at the top, the same scale Gemini 4 Hydrogen uses. Google recommends starting Neon at high and reserving xhigh for hard problems because it increases latency and token use.
........
Setting | Gemini 4 Neon | Gemini 4 Carbon |
Context window | 1,048,576 tokens, shared | 1,048,576 tokens, shared |
Synchronous output per hop | 65,536 (streaming and non-streaming) | 65,536 streaming, 32,768 non-streaming (Cloud docs) |
Long decode | Background mode, up to 1M minus input | Background mode, up to 1M minus input |
Thinking levels | low, medium, high, xhigh | low, medium, high |
Default thinking level | high | medium |
minimal / thinking_budget | Rejected | Rejected (HTTP 400) |
temperature, top_p, top_k | Remove; ignored in range | Remove; ignored |
Explicit caching | Not supported | Not supported |
Implicit caching | Prompts of 4,096+ tokens, 95% discount | Prompts of 4,096+ tokens, 95% discount |
Context in Antigravity | Up to 512K tokens | Not stated |
........
The default level shapes the bill as much as the list price. A team that moves from Carbon to Neon without setting thinking_level goes from medium to high by default, and Artificial Analysis recorded 123 million output tokens for Neon at high to run its Intelligence Index against 62.9 million at medium. Pinning the level explicitly is the only way to compare costs on equal terms.
··········
GEMINI 4 NEON AND CARBON ACCESS BY PLAN, APP AND CLOUD PLATFORM.
Access is where the two models diverge most for consumers. Google's Neon announcement gives the model to Google AI Pro and Ultra subscribers in the Gemini app, AI Studio and Antigravity, while Carbon's November 10 rollout to the consumer apps was announced for Google AI Ultra subscribers. Google AI Pro costs $19.99 a month; Ultra starts at $99.99 a month, with a $199.99 option that Google describes as 20 times the usage limits of Pro.
........
Surface | Gemini 4 Neon | Gemini 4 Carbon |
Gemini API | Paid tier only | Paid tier only |
Google AI Studio | Google AI Pro and Ultra subscribers | Google AI Ultra subscribers |
Gemini app | Pro and Ultra, model dropdown; Tasks can run on Neon | Ultra ("highest access" on the plan page) |
Google Antigravity | Pro (paid, or trial on eligible devices) and Ultra; default model for new conversations | Ultra |
Orchestrator (Neon) with Hydrogen | Pro and Ultra; not in Gemini for business | Not applicable |
AI Mode in Search | Not stated | Ultra |
Gemini for business | API and model dropdown | API, model dropdown, Antigravity, CodeMender |
Google Cloud Agent Platform | GA; global, US and EU | GA; verified enterprise customers in select regions |
Fairwind cyber variant | Eligible members, rolling access | Eligible members, rolling access |
Free and Google AI Plus | No access | No access |
........
On Google Cloud both models support batch inference, Provisioned Throughput, Standard and Priority PayGo, grounding with Google Search and Google Maps, code execution, function calling, URL context and computer use, and neither supports tuning, RAG Engine or the Live API. Carbon's Cloud model page adds a condition Neon's page does not carry, stating that Carbon is "available to verified enterprise customers in select regions". The Antigravity team's announcement adds that Orchestrator (Neon) pairs the two newest models in one conversation, with Neon planning and reviewing while Hydrogen edits code, updates packages and runs tests.
··········
HOW GOOGLE, ARTIFICIAL ANALYSIS AND VALS AI MEASURED NEON AND CARBON.
The benchmark figures come from three kinds of sources, and they are not interchangeable. Google's Neon evaluation document states that its scores are pass@1 with no majority voting, run through the Gemini API at the highest thinking setting unless noted, which for Neon means xhigh; FrontierCode 2.0 is the stated exception, run at high. Several of Google's scores are self-computed: Agents' Last Exam v1.1 on 122 tasks with Antigravity-CLI 1.4.8 and a five-hour window, Terminal-Bench Science 0.1 with a mini-swe-agent vision setup averaged over five attempts, and SEC-bench Pro on a 183-task May 2026 JavaScript engine split averaged over three attempts. Carbon's figures come from Google's earlier Carbon table, whose highest available setting is high.
Artificial Analysis tests each model at each effort level on the same scaffold and publishes cost and time per task, which makes it the most useful source for like-for-like work: Neon at xhigh, high, medium and low against Carbon at high, medium and low. Vals AI publishes one configuration per model, with Neon at xhigh and Carbon at high, both with a 1,048,576-token output ceiling and default sampling. Vals also shows inconsistencies on Carbon's page, where an October 28 update reports figures that differ from the current table; this report uses the current table values.
··········
GEMINI 4 NEON VS CARBON SCORES ON THE SAME GOOGLE BENCHMARKS.
The following benchmarks appear with the same name and version in both Google's Neon launch material and Google's Carbon results table. Neon's figures are at xhigh unless noted; Carbon's are at its highest setting. Settings for every Carbon row were not published individually, so small gaps should be read with caution.
........
Benchmark (Google-published) | Gemini 4 Neon | Gemini 4 Carbon | Higher score |
Agents' Last Exam v1.1 | 75.8% | 77.4% | Carbon |
AutomationBench (Zapier) | 56.3% | 53.9% | Neon |
Vending-Bench 2 (bank balance) | $16,843 | $16,484 | Neon |
Chartography | 73.5% | 72.5% | Neon |
RiemannBench | 84.0% | 72.0% | Neon |
SEC-bench Pro | 90.5% | 88.2% | Neon |
SWE-sweep | 12.8% (mini-SWE setup); 39.0% (Antigravity) | 9.6% (setup not stated) | Neon, setups differ |
Artificial Analysis Intelligence Index | 61 (xhigh) | 59 (v5.0, at Carbon launch) | Neon |
........
The largest gap in Google's own figures is RiemannBench, a mathematics benchmark, where Neon's 84.0% is twelve points above Carbon's 72.0% and also above the 76.0% Google reported for Argon. Agents' Last Exam v1.1 is the one shared Google benchmark where Carbon stays ahead, by 1.6 points. Google also states, without publishing per-model token counts, that Neon matches frontier performance on ExploitGym using half of Carbon's tokens and produces more secure code than Carbon at every AutoBaxBench difficulty level.
··········
BENCHMARKS GOOGLE PUBLISHED FOR ONLY ONE OF THE TWO GEMINI 4 MODELS.
Many headline numbers cannot be compared directly because Google reported them for one model only or on different benchmark versions. The Neon launch used Terminal-Bench 4.1, FrontierCode 2.0 and CWE-bench v2, while Carbon's table used Terminal-Bench 4.0, FrontierSWE 2 and the first version of CWE-bench.
........
Benchmark | Gemini 4 Neon | Gemini 4 Carbon | Comparable |
Terminal-Bench 4.1 (cheating-adjusted) | 65.2% | Not published | No |
Terminal-Bench 4.0 (Artificial Analysis) | See independent table | 65.2% | No |
FrontierCode 2.0 (high) | 46.2% | Not published | No |
FrontierSWE 2 | Not published | 61.2% | No |
CWE-bench v2 (Collinear) | 70.7%, a 3.4% uplift over Carbon per Google | Not published | Google states the delta only |
CWE-bench (v1) | Not published | 71.0% | No |
WebExploitBench v1.2 | 51.26% | Not published | No |
AutoBaxBench, hard tier | 76% | Lower, figure not published | Google states the direction only |
ExploitGym | Not published | 54.8% | No |
OSWorld-2.1 (offline subset) | Not published | 81.5% | No |
GraphWalks, 256K to 1M (BFS F1) | Not published | 96.2% | No |
Blueprint-Bench 2 | Not published | 58.6% | No |
Agent Arena Index | 1336, second | Not published | No |
WebDev Arena | Not published | 1768 Elo, fourth | No |
........
The missing figures weigh most on long-context retrieval and computer use. Carbon's 96.2% on GraphWalks across 256K to 1M tokens and 81.5% on OSWorld-2.1 have no Neon equivalent, so buyers whose workloads depend on reading near the full million-token window or on desktop automation have evidence only for Carbon.
··········
INDEPENDENT EVALUATIONS OF GEMINI 4 NEON AND CARBON BY ARTIFICIAL ANALYSIS AND VALS AI.
Artificial Analysis published its comparison and an article on November 18. Its Intelligence Index puts Neon at xhigh one point above Carbon at high, and Neon at high level with Carbon at high. The article ranked Neon third behind GPT-6.2 Astra (max) and Claude Fable 5.2 (max); the model page listed it sixth of 96 on November 19. On the Coding Agent Index, run in the Antigravity CLI, Neon at xhigh scored 59 against 55 for Carbon at high.
........
Artificial Analysis metric | Neon xhigh | Neon high | Carbon high | Carbon medium |
Intelligence Index | 61 | 60 | 60 | 58 |
Terminal-Bench 4.0 | 63.6% | 61.6% | 65.2% | 60.6% |
FrontierCode-AA 2.0 | 44.4% | 43.3% | 40.6% | 39.5% |
HLE-Diamond | 71.3% | 68.4% | 69.1% | 67.4% |
Terminal-Bench-Science 0.1 | 70.0% | 65.7% | 64.8% | 62.4% |
CritPt v2 (hard subset) | 60.5% | 59.0% | 69.0% | 69.0% |
GDP.pdf | 28.0% | 33.4% | 28.4% | 29.6% |
AutomationBench-AA | 79.3% | 79.2% | 77.8% | 77.6% |
AA-Omniscience v1.1 | 55 | 54 | 54 | 47 |
AA-Briefcase v1.1 (Elo) | 1700 | 1632 | 1611 | 1602 |
Cost per task | $6.44 | $3.74 | $7.76 | $6.59 |
Time per task | 21.5 min | 14.8 min | 37.2 min | 29.6 min |
........
Artificial Analysis priced Carbon at its standard $4/$20 rate, not the introductory one, and noted that Neon uses about 15% more output tokens per task than Carbon at the top settings, so Neon's lower cost per task comes from its lower token prices. CritPt v2 is Carbon's clearest independent lead, at 69.0% against 60.5%, and Terminal-Bench 4.0 is the other measurement where Carbon at high beats every Neon setting.
........
Vals AI benchmark | Gemini 4 Neon (xhigh) | Gemini 4 Carbon (high) |
Vals Index | 70.87%, rank 2/42 | 70.45%, rank 4/42 |
Vals Finance Index | 68.31%, rank 1/45 | 67.05%, rank 2/45 |
Vals Legal Index | 78.36%, rank 1/43 | 75.93%, rank 4/43 |
Legal Research Bench | 56.25%, rank 1/99 | 55.77%, rank 2/99 |
Tax Agent Bench | 54.89%, rank 1/90 | 51.78%, rank 3/90 |
Finance Agent v2.1 | 68.56% | 67.16% |
Vibe Code Bench v1.1 | 93.70%, rank 2/133 | 89.06%, rank 11/133 |
Terminal-Bench 4.0 | 61.62% | 63.13% |
HLAB-Vals-1.1 | 19.58% | 17.50% |
EMB v1.1 | 76.46% | 76.47% |
Vals Index cost per test | $7.405 | $10.51 |
Latency | 31 min 25 s | 47 min 45 s |
........
The Vals results line up with Artificial Analysis on direction. Neon leads the finance, legal and tax benchmarks, ties on EMB, and trails Carbon on Terminal-Bench 4.0, while costing about 30% less per Vals Index test and finishing in roughly two thirds of the time.
··········
GEMINI 4 NEON AND CARBON SPEED, LATENCY AND OUTPUT-TOKEN USE.
Speed is Neon's most consistent advantage in the independent data. Artificial Analysis measured Neon at 97 to 110 output tokens per second across its four levels, against 47 to 58 for Carbon, roughly double. Time to first answer token depends heavily on the thinking level: Neon at high took 40.64 seconds and Carbon at high 73.04 seconds, while Carbon at low answered first token in 9.00 seconds against 13.93 for Neon at low. End-to-end time on Artificial Analysis's standard prompt was 81.29 seconds for Neon at high against 181.16 seconds for Carbon at high.
Token consumption rises steeply with Neon's effort level. Artificial Analysis recorded 283 million output tokens to run its Intelligence Index at xhigh, 123 million at high, 62.9 million at medium and 28.3 million at low. It did not publish the equivalent figure for Carbon on the pages checked. For latency-sensitive applications, the practical reading is that Neon at medium (8.3 minutes per task, $1.90) or high (14.8 minutes, $3.74) delivers Carbon-medium-level or Carbon-high-level scores at a fraction of Carbon's time.
··········
COST OF THREE TYPICAL WORKLOADS ON GEMINI 4 NEON AND CARBON.
The examples below use Gemini API list prices and apply the same token counts to both models, which isolates the price difference. Artificial Analysis's measurement that Neon used about 15% more output tokens than Carbon at their top settings is shown as a sensitivity line for the coding case. The coding agent assumes 40 turns per task with 60,000 input tokens per turn, 85% served from the implicit cache, and 3,000 output tokens per turn including thinking. The document review assumes 500 documents a month at 300,000 uncached input tokens and 8,000 output tokens each. The support assistant assumes 100,000 conversations a month at 6,000 input tokens (70% cached) and 600 output tokens.
........
Workload | Gemini 4 Neon | Carbon to Nov 30 | Carbon from Dec 1 |
Coding agent, one task | $3.19 | $2.12 | $4.25 |
Coding agent, one task, Neon with 15% more output | $3.46 | $2.12 | $4.25 |
Long-document review, 500 a month | $510.00 | $340.00 | $680.00 |
Support assistant, 100,000 conversations a month | $1,503.00 | $1,002.00 | $2,004.00 |
........
At equal token counts Neon costs exactly 75% of Carbon's December rate and 150% of Carbon's November rate, because every price component keeps the same ratio. The independent per-task figures widen Neon's advantage once thinking levels differ: Artificial Analysis found Neon at high matching Carbon at high on its Intelligence Index for $3.74 against $7.76 per task, a 52% saving. For batch work on Google Cloud, Neon's $1.50/$7.50 global batch rate sits between Carbon's $1.00/$5.00 promotional batch rate and its $2.00/$10.00 rate from December.
··········
SAFETY RESULTS AND CYBER SAFEGUARDS FOR GEMINI 4 NEON AND CARBON.
Google states that Neon runs on the same safety and security stack as Carbon, with comparable real-time cyber and biosecurity safeguards. The Neon model card carries forward Frontier Safety Framework results from Gemini 4 Argon, explaining that Neon's capability increase over Argon fell below the Material Capability Change threshold, so no full critical capability evaluation was run. Carbon's model card does not report a status for each domain and refers readers to a separate Safety and Security report.
........
Safety measure | Gemini 4 Neon | Gemini 4 Carbon |
Frontier Safety Framework status | Argon results carried forward: biological CBRN and cybersecurity CCL cannot be ruled out; harmful manipulation and ML R&D CCLs not reached | Not stated in model card; separate safety report |
Terminal-Bench 4.0 successful cheating | 0.0% | 0.0% |
Gray Swan indirect prompt injection | 0.0% attack success at highest level with safeguards | 0.0% attack success |
Under-service vs Carbon | 43% to 56% lower | Baseline |
Overreach vs Carbon | 15% to 35% lower, by effort level | Baseline |
Coding shortcut rate | 2.45%, lowest of the Gemini 4 models | Higher, figure not published |
Misaligned action rate (agentic safety suite) | 0.09% | Not published |
Cyberattack red-teaming | Not published | 99% of attack trajectories blocked; 0.8% per-attempt breach rate |
Refusal rate on Terminal-Bench 4.0 | 1.98% | Not published |
........
Both models are offered in a cyber variant to verified defenders through the Fairwind Program, and Google says cybersecurity benchmarks were evaluated with safety filters off. Neon's model card compares its content-safety evaluations with Hydrogen, not Carbon, and reports that all differences fall within confidence intervals, with unjustified refusals down 3.00 percentage points.
··········
WHAT GOOGLE HAS NOT PUBLISHED ABOUT GEMINI 4 NEON AND CARBON.
Several figures that would complete the comparison are missing from official pages as of November 19, 2026. The Gemini API developer page for Neon gives no batch or Flex price, although Google Cloud lists Neon batch pricing. Neither model's developer page gives numeric rate limits, and the Antigravity announcement gives no quota figures beyond saying Neon's lower cost per task makes quota last longer. Google has not published Neon scores for OSWorld-2.1, GraphWalks, Blueprint-Bench 2 or ExploitGym, and it has not published Carbon scores on Terminal-Bench 4.1, FrontierCode 2.0 or CWE-bench v2, which leaves the coding and cyber comparisons dependent on Artificial Analysis and Vals AI.
The model cards also leave the architecture, parameter counts and training compute undisclosed for both models, and Google has not said how long Carbon will stay available alongside Neon. Artificial Analysis did not publish Carbon's total output-token count for its index on the pages checked, so the 15% token gap reported in its article cannot be reproduced at every effort level from public data.
··········
WHICH TO CHOOSE BETWEEN GEMINI 4 NEON AND GEMINI 4 CARBON BY WORKLOAD.
The choice follows the price calendar and the benchmark pattern. Before December 1, Carbon is the cheaper model per token on the API; from December 1, Neon is 25% cheaper per token and usually cheaper per task by a wider margin because it reaches Carbon's scores at lower thinking levels.
........
Workload or user | Recommended model | Reason |
Coding agents and repository work | Gemini 4 Neon | Coding Agent Index 59 vs 55; FrontierCode-AA 44.4% vs 40.6%; Vibe Code Bench 93.70% vs 89.06% |
Terminal-heavy automation | Gemini 4 Carbon | Leads Terminal-Bench 4.0 in both Artificial Analysis and Vals AI tests |
Finance, legal and tax analysis | Gemini 4 Neon | First on Vals Finance, Legal, Legal Research and Tax Agent benchmarks |
Hard physics and research reasoning | Gemini 4 Carbon | CritPt v2 69.0% vs 60.5% |
Mathematics | Gemini 4 Neon | RiemannBench 84.0% vs 72.0% in Google's figures |
Retrieval across 256K to 1M tokens | Gemini 4 Carbon | Only model with published GraphWalks results (96.2%) |
Latency-sensitive agents | Gemini 4 Neon | About twice the output speed; 14.8 vs 37.2 minutes per task at high |
Heavy API spend before Nov 30 | Gemini 4 Carbon | $2/$10 introductory rate vs Neon's $3/$15 |
Google AI Pro subscribers | Gemini 4 Neon | Included in the Gemini app, AI Studio and Antigravity on Pro |
Google Cloud teams without verification | Gemini 4 Neon | Carbon's Cloud page limits it to verified enterprise customers in select regions |
........
For most developers the decision comes down to the December 1 price change. A team running coding agents, document analysis or finance workflows on Carbon will pay 25% less per token on Neon from that date and, on the Artificial Analysis and Vals AI measurements, will get equal or higher scores in less time; teams whose work depends on Terminal-Bench-style shell automation, CritPt-level physics or near-million-token retrieval have published evidence only for keeping Carbon. Readers comparing the earlier pair in the family can find our Gemini 4 Carbon vs Gemini 4 Argon comparison.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org




