Gemini 4 Carbon: A Second-Place MKA Result For A Sixth Of The Cost
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Gemini 4 Carbon: A Second-Place MKA Result For A Sixth Of The Cost on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get hardware and tech essentials delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Google made Gemini 4 Carbon available to paying developers, businesses and Google AI Ultra subscribers on November 10, 2026. Carbon placed second behind Claude Fable 5.2 on the new MKA benchmark, while Google DeepMind reported a lower cost per task; the benchmark was co-developed by Google and its tasks are private.

Google opened Gemini 4 Carbon to paying developers, businesses and Google AI Ultra subscribers on November 10, the same day Google DeepMind and Artificial Analysis introduced the MKA benchmark. Carbon scored 54.3%, placing second behind Anthropic’s Claude Fable 5.2 at 56.6%; Google DeepMind’s report put Carbon’s cost at $5.63 per task, compared with $35.28 for Fable 5.2.

The MKA results cover 200 expert-built tasks across 10 sectors. Claude Opus 5.5 scored 53.0%, putting the top three close together. The source report says those scores are within the margin of error, so the narrow ranking differences should not be treated as definitive evidence of a performance gap. Google DeepMind reported the per-task costs; they are benchmark-specific figures, not a complete measure of what a customer would pay across other workloads.

Carbon is available through the paid Gemini API and Google AI Studio, as well as Gemini for business. In the Gemini app and AI Mode in Search, access is limited to Google AI Ultra subscribers. Google’s listed API price is $2 per million input tokens and $10 per million output tokens through November 30, rising to $4 and $20, respectively, on December 1, 2026.

Google’s model card reports that Carbon leads on 15 of 17 benchmarks selected by Google. Those comparisons were marked as current through October 2026 and do not include Claude Fable 5.2, which the source says was released November 9. On Google’s selected tests, GPT-6 Astra scored higher on FrontierSWE 2, while Claude Opus 5.5 led on OSWorld-2.1.

At a glance
reportWhen: Access opened November 10, 2026; pricin…
The developmentGoogle opened Gemini 4 Carbon to paying users as a new benchmark ranked it second behind Claude Fable 5.2, with a much lower reported cost per task.
Crypto market snapshot
Fear & Greed Index
29/100 — Fear
Bitcoin BTC$73,973▼ 0.5%
Ethereum ETH$2,055▼ 1.7%
Tether USDT$0.9993▲ 0.0%
BNB BNB$680.93▼ 1.0%
USDC USDC$0.9997▲ 0.0%
XRP XRP$1.06▼ 2.0%
Solana SOL$83.86▼ 0.8%
TRON TRX$0.3171▼ 0.5%
Live data · CoinGecko · alternative.me (24h change)
AI Models · Briefing

Gemini 4 Carbon: A Second-Place MKA Result for a Sixth of the Cost

Google’s model ranked second on MKA, with a much lower reported cost per task than the leader. The close scores and private benchmark tasks limit what the ranking can prove.

$5.63
Carbon cost per MKA task
54.3%
Carbon’s MKA score
$35.28
Fable 5.2 cost per MKA task
200
Expert-built MKA tasks

A close top three

The source report says the score differences fall within the margin of error.

Claude Fable 5.256.6%
Gemini 4 Carbon54.3%
Claude Opus 5.553.0%

What the cost comparison means

The reported gap may help businesses choose what to test, but it applies to this benchmark’s accounting.

Potential advantage
  • Carbon’s reported MKA task cost is lower.
  • Could matter if MKA reflects a team’s own work.
What to check
  • Token use, task design and model settings affect cost.
  • Google reports higher token use and occasional slowness or timeouts.

Access and API pricing

Paid access opened November 10, 2026; app and Search access is limited to Google AI Ultra subscribers.

API rateThrough November 30From December 1
Input, per million tokens$2$4
Output, per million tokens$10$20

Read the benchmark with care

MKA was co-developed by Google DeepMind, and its tasks are private, limiting independent inspection and reproduction.

  • The top three scores are reported within the margin of error.
  • Benchmark task costs do not guarantee costs on customer workloads.
  • Google’s selected benchmark comparisons predate Fable 5.2’s release.
For teamsTest Carbon on your own tasks and evaluate cost, speed and reliability together.
The takeaway
Carbon pairs a close second-place MKA score with a much lower reported task cost, but real-world value depends on independent evidence and your workload.

Cost and Performance on MKA

The result gives businesses a fresh comparison between benchmark performance and reported task cost. Carbon’s second-place score was close to Fable 5.2’s, while the reported MKA cost was substantially lower. If the benchmark reflects a company’s own work, that cost difference could affect which model it tests or deploys. It does not establish that Carbon will be cheaper for every task: token use, task design, model settings and the benchmark’s accounting method all matter.

The launch also illustrates why a single leaderboard position is an incomplete guide to model choice. Carbon is second on MKA, but the source reports a different ordering on Artificial Analysis’ Intelligence Index: Carbon scores 60, compared with Fable 5.2 at 65 at its maximum setting and 62 at “high.” That index and MKA assess different things. The practical question for users is whether a model’s results, speed and price suit their particular work, not which model holds one rank.

Google says Carbon is intended for difficult, multi-step coding and open-ended scientific research. Its model card also describes higher token use and occasional slowness or timeouts. Those limitations may matter to teams weighing the lower MKA cost against latency, reliability and total usage. The benchmark provides a comparison point, not a guarantee of performance in production.

Products Worth Considering

LG 27G610A-B 27-inch UltraGear QHD (2560 x 1440) IPS Gaming Monitor, 200Hz, 1ms, AMD FreeSync Premium, VESA DisplayHDR 400, HDMI, DisplayPor

LG 27G610A-B 27-inch UltraGear QHD (2560 x 1440) IPS Gaming Monitor, 200Hz, 1ms, AMD FreeSync Premium, VESA DisplayHDR 400, HDMI, DisplayPor

LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDR True Black 400

LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDR True Black 400

LG UltraGear 32GX870A-B 32-inch 4K OLED Dual-Mode Gaming Monitor

LG UltraGear 32GX870A-B 32-inch 4K OLED Dual-Mode Gaming Monitor

As an affiliate, we earn on qualifying purchases.

A Specialist Model Goes Paid

Carbon is built on Google’s Gemini 4 Argon and is designed to spend more time reasoning and verifying answers. Google describes it as suited to tasks that test a model’s limits. The company says Argon remains the better choice for most software engineering and knowledge work, framing Carbon as a specialist rather than a universal replacement.

According to Google’s model card and developer guide, Carbon accepts text, images, audio and video, and produces text. Its context window supports up to one million tokens for input and output combined. It offers low, medium and high thinking settings, with medium as the default. Google lists the model’s knowledge cutoff as March 2026, while cautioning that knowledge may be less current in some areas.

Before the wider paid release, Carbon was made available to a small group of cyber defenders through Google’s Fairwind Program. The source report says its security features remain restricted to verified defenders. Artificial Analysis lists October 28 as the model’s release date; Google opened broader paid access on November 10.

““tasks that test the limits of a model’s intelligence””

— Google, in the Gemini 4 Carbon model card

Limits of the MKA Comparison

MKA was co-developed by Google DeepMind, and its tasks are private, according to the source report. That means outside researchers cannot independently inspect the task set or reproduce the results from the materials described. The reported scores offer a comparison, but readers should account for Google’s role in creating the benchmark and the lack of public tasks.

The source says the top three scores fall within the margin of error, but does not provide the underlying uncertainty ranges or further methodological details here. It also does not explain fully how the reported cost per task was calculated, including how model settings and usage were standardized. The figures should not be assumed to predict costs for a company’s own requests.

Carbon’s broader performance is also still developing. Google’s model card uses tests and rivals selected by Google and predates Fable 5.2’s release, while Artificial Analysis uses a separate set of tests. Google lists hallucinations, occasional slowness or timeouts, and increased token use as limitations. How often these occur in real deployments is not established by the information provided.

Pricing Changes on December 1

Developers and businesses can test Carbon through Google’s paid developer and business services, while access through the Gemini app and Search’s AI Mode remains tied to Google AI Ultra. The listed API rates are promotional through November 30: from December 1, 2026, Google’s stated prices rise from $2 to $4 per million input tokens and from $10 to $20 per million output tokens.

Further comparisons may become possible as independent testing and real-world deployments produce more evidence. For now, the available information establishes Carbon’s MKA score, its reported benchmark cost and Google’s stated access conditions; it does not settle how Carbon will perform or price out across different customer workloads. Google’s stated limits, including possible timeouts and heavier token use, are factors developers will need to evaluate during testing.

Key Questions

What did Gemini 4 Carbon score on MKA?

Carbon scored 54.3% on the new benchmark. Claude Fable 5.2 scored 56.6%, and Claude Opus 5.5 scored 53.0%. The source report says the top three results are within the margin of error.

How much does Carbon cost on MKA compared with Fable 5.2?

Google DeepMind’s report gives a cost of $5.63 per MKA task for Carbon and $35.28 for Fable 5.2. These are benchmark-specific reported costs, not guaranteed prices for other tasks.

Who can use Gemini 4 Carbon?

Paying developers can access it through the Gemini API and Google AI Studio, and businesses can use it through Gemini for business. Access in the Gemini app and AI Mode in Search is limited to Google AI Ultra subscribers.

When does Carbon’s API price change?

Google lists rates of $2 per million input tokens and $10 per million output tokens through November 30, 2026. From December 1, 2026, the listed rates become $4 and $20, respectively.

Is Carbon the top-performing model overall?

No single result establishes an overall winner. Carbon placed second on MKA, and the source report says the leading MKA scores are within the margin of error. On Artificial Analysis’ Intelligence Index, Carbon scored 60, below Fable 5.2 at 65 on its maximum setting and 62 at “high.”

Source: ThorstenMeyerAI.com

Nothing in this article is financial or investment advice. Cryptocurrency and precious-metal investments carry significant risk — do your own research and consider a licensed advisor.
You May Also Like

A Loop Fix, Not An IQ Jump: GPT-6 Astra Edited Another Model

A blog post reports that GPT-6 Astra edited two Qwen3.5-0.8B layers, raising its MATH-500 score while reducing—but not eliminating—looping.

The Coding Singularity Is Real — and Steeper Than Clark Presented

Recent data confirms the coding singularity is unfolding faster than previously estimated, with AI capabilities surpassing earlier forecasts and deployment expanding.

Following Macron, Trump Joins Forces With PM Modi on Artificial Intelligence Initiatives

Strengthening US-India relations, Trump and Modi’s AI initiatives raise questions about global governance and economic impact. What could this mean for the future?

Vocal-strain load tracking for working singers

A new app prototype aims to monitor vocal load for professional singers, providing early warnings to prevent injury during touring schedules.