Perplexity says Google’s Gemini 4 Carbon scored 0.857 on its WANDR research benchmark at $5.86 per task, “matching Claude Fable 5.2 at 40% of the cost”. A day earlier, Perplexity posted that Fable 5.2 set “a record score of 0.851 at $14.51 per task”. That is $8.65 less per task, or 40.4% of Fable 5.2’s cost, by our arithmetic.
Perplexity made Carbon available in Perplexity Computer for Pro and Max subscribers on 10 November, according to its post on X. Google’s Carbon price is introductory: its pricing page shows it doubling on 1 December.
Business Pill · Usage-Based Pricing
An evergreen Business Engineer explainer on usage-based pricing: paying for work, not for seats. It teaches the general idea only and does not discuss Perplexity, Gemini 4 Carbon or Claude Fable 5.2.
The key insight: Perplexity reported on 10 November that Gemini 4 Carbon scored 0.857 on its WANDR benchmark at $5.86 per task, against 0.851 at $14.51 for Claude Fable 5.2 a day earlier. That is 40.4% of the cost, by our arithmetic, before Google’s list price for Carbon doubles on 1 December.
Perplexity’s post says Carbon is available “in Perplexity Computer for Pro and Max subscribers, and in Search for Max subscribers”. Chief executive Aravind Srinivas wrote that “Carbon is 2.5x cheaper than Fable 5.2 with similar capabilities on agentic research tasks”. By our arithmetic, $14.51 divided by $5.86 is 2.48 times.
The score gap is small: 0.857 against 0.851, a difference of 0.006 points, by our arithmetic. Across 1,000 tasks, the reported costs come to $5,860 for Carbon and $14,510 for Fable 5.2.
Google’s own launch post made a similar claim. It says Carbon “ranks #1 on CursorBench 4.0, Perplexity’s WANDR, and Zapier’s AutomationBench at more than 2X lower cost per task compared to the next best external model”. We covered the Carbon launch and its price schedule earlier today.
Cost per WANDR task as reported by Perplexity: Gemini 4 Carbon $5.86 (score 0.857) and Claude Fable 5.2 $14.51 (score 0.851). The $11.72 bar is our arithmetic ($5.86 x 2) if Perplexity’s cost scaled with Google’s list price, which doubles on 1 December 2026. Source: Perplexity posts on X, 9 and 10 Nov 2026; Gemini API pricing page.
What Was Posted on X
Gemini 4 Carbon is now available in Perplexity Computer for Pro and Max subscribers, and in Search for Max subscribers.
It scores 0.857 on our WANDR benchmark at $5.86 per task, matching Claude Fable 5.2 at 40% of the cost.
Perplexity posted on 10 November that Gemini 4 Carbon is available in Perplexity Computer and scores 0.857 on its WANDR benchmark at $5.86 per task, matching Claude Fable 5.2 at 40% of the cost. This piece is built on these figures.
Claude Fable 5.2 is now available in Perplexity Computer for Pro and Max subscribers, and in Search for Max subscribers.
On our WANDR benchmark, it achieved Pareto frontier of performance and efficiency, with a record score of 0.851 at $14.51 per task.
A day earlier, Perplexity posted that Claude Fable 5.2 set a record WANDR score of 0.851 at $14.51 per task. The piece compares these two posts.
Rolling out Gemini 4 Carbon to all Perplexity Pro and Max subscribers. You can create persistent Teammates with them or use it for one off agentic sessions in Computer. Carbon is 2.5x cheaper than Fable 5.2 with similar capabilities on agentic research tasks.
Perplexity chief executive Aravind Srinivas wrote that Carbon is 2.5x cheaper than Fable 5.2 with similar capabilities on agentic research tasks.
๐จ GOOGLE ROARS BACK WITH CARBON TO OVERTAKE ANTHROPIC
– Carbon outperforms Opus 5.5 – 250% cheaper than Opus – 500% cheaper than Fable – truly frontier class – not benchmark maxxed! – It totally works in the real world.
The only drawback, and it’s a significant one, is that it is about 3x slower than Opus even at low effort.
If they manage to fix it and keep the discount, they will OVERTAKE Anthropic! Available today on ChatLLM ๐
Bindu Reddy, chief executive of Abacus.AI, posted on 11 November that Gemini 4 Carbon outperforms Claude Opus 5.5, is 250% cheaper than Opus and 500% cheaper than Fable, and is about 3x slower than Opus even at low effort; she wrote that Google would overtake Anthropic if it fixes the speed and keeps the discount. Those claims are hers. The update box further down sets them against the LiveBench leaderboard, where Carbon scores 83.5 overall against 83.4 for Opus 5.5 and 84.9 for Fable 5.2.
What WANDR Measures
Perplexity released WANDR in July as “an open benchmark and evaluation harness built around 500 realistic, challenging data-collection tasks for knowledge work”. The paper says WANDR asks whether an agent “can build a large collection and back up every member with specific evidence”, for jobs such as competitive mapping, due diligence, literature review and market analysis.
In that July paper, scores were reported as soft and hard F1, and the strongest system reached 0.363 soft F1. Perplexity’s posts this week do not say which scoring the 0.857 and 0.851 figures use. We covered Perplexity’s Agent API effort levels on 6 November, when its ultra level cost $21.71 per WANDR task.
The Price Change on December 1
Google lists Carbon at $2 per million input tokens and $10 per million output tokens through 30 November, then $4 and $20 from 1 December. Anthropic lists Fable 5.2 at $8 and $40. Carbon’s introductory token prices are 25% of Fable 5.2’s; December’s are 50%.
Perplexity’s cost per task, 40.4% of Fable 5.2’s, is higher than the 25% token-price ratio. If Perplexity paid list prices, that would mean Carbon used about 1.6 times as many tokens per task at a similar input and output mix, by our arithmetic. Google’s post says Carbon uses “more tokens than Argon to explore thoroughly, reason deeply, and run extended verification loops”.
If Perplexity’s cost scaled with Google’s list price, Carbon’s $5.86 would become $11.72 per task in December, about 81% of Fable 5.2’s $14.51, by our arithmetic. Google’s Cloud catalog already listed Carbon at $4 and $20 this morning.
Update (10 Nov): Carbon reaches GitHub Copilot.
GitHub’s changelog, dated 9 November, says Gemini 4 Carbon “is now generally available in GitHub Copilot” and that the model “is billed at provider list pricing under usage-based billing”. GitHub’s post on X repeated the news on 10 November.
The changelog lists Copilot Pro+, Max, Business and Enterprise users, and eight places to pick the model: Visual Studio Code, Visual Studio, Copilot CLI, the Copilot coding agent, the Copilot app, JetBrains IDEs, Xcode and Eclipse. It says “Rollout will be gradual”, and that under default model enablement new models are automatically enabled unless an administrator turns them off.
As we read it, billing at provider list pricing means Copilot users who choose Carbon pay Google’s list rate, which the section above shows doubling on 1 December. The changelog itself gives no per-request price or multiplier.
Update (11 Nov): LiveBench puts Carbon level with Opus 5.5, not ahead of Fable.
On the LiveBench leaderboard, last updated 11 November, Gemini 4 Carbon scores 83.5 overall against 83.4 for Claude Opus 5.5, a gap of 0.1 points by our arithmetic. Claude Fable 5.2 (xhigh) leads at 84.9, ahead of GPT-6 Astra at 84.5 and Claude Fable 5.1 at 84.0, so Carbon is fourth of the rows shown.
The cost column is where Carbon stands out: $2.024 per successful task, against $6.723 for Opus 5.5 and $9.850 for Fable 5.2. That is 3.3x and 4.9x cheaper, by our arithmetic. By category, Carbon leads Opus 5.5 in coding (91.0 vs 87.1) and research (92.5 vs 88.9), and trails it in design (80.2 vs 83.3) and browser use (80.0 vs 88.4).
LiveBench’s page says its results power an agent “available via an Abacus AI subscription”. Bindu Reddy, Abacus.AI’s chief executive, posted that Carbon “outperforms Opus 5.5” and that Google “will OVERTAKE Anthropic” if it fixes speed and keeps the discount; that framing is hers. She also wrote Carbon is “about 3x slower than Opus even at low effort”. LiveBench’s own column shows 45 minutes per task for Carbon against 50 for Opus 5.5, so the slowdown she describes is not visible on the leaderboard.
The Structural Read
As we read the posts, the comparison Perplexity is making is cost per completed research task, not price per token. On that measure Carbon’s advantage is smaller than its token prices suggest, and it would narrow further in December if Perplexity’s costs follow Google’s list price.
Aravind Srinivas, Perplexity chief executive, on X
“Carbon is 2.5x cheaper than Fable 5.2 with similar capabilities on agentic research tasks.”
Three Implications
FOR TEAMS BUYING RESEARCH AGENTS Perplexity reports Carbon at $5.86 and Fable 5.2 at $14.51 per WANDR task; across 1,000 tasks that is $5,860 against $14,510, by our arithmetic.
FOR DECEMBER BUDGETSGoogle lists Carbon at $2 and $10 per million input and output tokens through 30 November and $4 and $20 from 1 December; Anthropic lists Fable 5.2 at $8 and $40.
FOR READING THE SCORES WANDR is Perplexity’s own benchmark; its July paper reported soft and hard F1, and this week’s posts do not say which scoring the 0.857 and 0.851 figures use.
The framework says: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”
As we read it, Perplexity’s figures are a measure of that marginal cost: what one research task costs on each model. The framework does not discuss Perplexity, WANDR or Gemini 4 Carbon.
What Is Not Established
WANDR is Perplexity’s own benchmark, and the scores and costs have not been independently checked. Perplexity has not said what it pays Google or Anthropic, whether its costs include caching or tools, or which WANDR scoring the posted figures use. The December estimate assumes list prices and the same token use; it is not a Perplexity figure.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Perplexity says Gemini 4 Carbon scored 0.857 on WANDR at $5.86 per task, against Claude Fable 5.2’s 0.851 at $14.51: 40.4% of the cost, by our arithmetic. If that cost tracked Google’s list price, which doubles on 1 December, it would be about $11.72, still below Fable 5.2.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Scores, costs and quotations come from Perplexity’s posts on X of 9 and 10 November 2026 and its chief executive’s post; Carbon prices come from Google’s launch post and Gemini API pricing page, Fable 5.2 prices from Anthropic’s pricing page, read 10 November 2026. The 40.4%, 2.48 times, $8.65, 1.6 times and $11.72 figures are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
