# Google Opens Gemini 4 Carbon at $2 per Million Input Tokens, a Quarter of Anthropic's Fable 5.2 Price

Cursor reports Carbon scoring 60.4% on CursorBench against 60.0% for Fable 5.2 Max, but the introductory rate doubles on December 1, and Google's claim of lower cost per task is not backed by published data.

- Published: 2026-11-11T06:08:31.146Z
- Canonical: https://polylog.news/ai/2026-11-11/google-opens-gemini-4-carbon-at-2-per-million-input-tokens-a
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Google DeepMind](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/), [Polylog editors](https://polylog.news), [Cursor Docs](https://cursor.com/docs/models/gemini-4-carbon), [Cursor on X](https://x.com/cursor_ai/status/2120193584670580857), [MLQ](https://mlq.ai/news/google-makes-gemini-4-carbon-widely-available-at-2-per-million-input-tokens-and-previews-agent-swarms/), [arXiv cs.LG](https://arxiv.org/abs/2611.12956)

Google [made Gemini 4 Carbon available](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/) to paying developers on November 10. The general release comes two weeks after the October 28 announcement, when Google [limited access](https://syntaxandsignal.tech/blog/gemini-4-carbon-google-deepmind-intelligence-index-cybersecurity-fairwind-2026/) to selected cyber defenders and partners in its Fairwind program. According to [MLQ](https://mlq.ai/news/google-makes-gemini-4-carbon-widely-available-at-2-per-million-input-tokens-and-previews-agent-swarms/), access now covers the Gemini API, enterprise customers, Google AI Ultra subscribers in the Gemini app, AI Mode in Search, and Google's Antigravity coding environment. Pro and free consumer tiers are excluded.

Pricing matters most to companies that build products on top of the model. [Cursor's model documentation](https://cursor.com/docs/models/gemini-4-carbon) lists Google's introductory rate of $2 per million input tokens and $10 per million output tokens through November 30. From December 1 the rate rises to $4 input, $0.20 cached read and $20 output. One day earlier, Anthropic [cut Claude Fable 5.2](https://www.implicator.ai/anthropic-fable-5-2-mythos-5-2-price-cut/) to $8 input and $40 output. Carbon therefore costs one quarter of Fable per token at the introductory rate and one half at the standard rate.

The capability evidence is close and mostly comes from vendors. Cursor [reported](https://x.com/cursor_ai/status/2120193584670580857) that Carbon at high effort scored 60.4% on CursorBench, the company's internal coding benchmark, against 60.0% for Fable 5.2 Max. The Russian-language channel [AI ML Big Data](https://t.me/ai_machinelearning_big_data/11332) summarized Google's claim that Carbon ranks first on three agentic benchmarks: CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench. Anthropic's own launch table [shows](https://officechai.com/ai/fable-5-2-benchmarks/) Carbon ahead on AutomationBench at 53.9% to 51.4%. The same table shows Carbon behind on FrontierCode 2.0, at 41.1% against Fable's 49.1%. Artificial Analysis [scored Carbon 59](https://syntaxandsignal.tech/blog/gemini-4-carbon-google-deepmind-intelligence-index-cybersecurity-fairwind-2026/) on its Intelligence Index, and Fable 5.2 later [scored 65](https://the-decoder.com/claude-fable-5-2-and-mythos-5-2-cement-anthropics-ai-lead-just-before-its-ipo/). A 0.4-point lead on one benchmark run by a vendor does not establish a ranking. Cursor also has not published the per-task cost data that would support Google's claim that Carbon costs less than half as much per completed task.

Google also previewed "swarms" in Antigravity, where one prompt launches a team of agents. [AlphaCorp](https://alphacorp.ai/blog/gemini-4-carbon-launch-benchmarks-pricing-and-everything-you-need-to-know) warns that swarms use quota much faster than a single chat session. That makes the per-token rate less useful than cost per completed task as a basis for procurement.

Google has not disclosed Carbon's architecture or how it achieves the price. Research posted on November 11 shows one way labs reduce long-context serving cost. The [ADIndex paper](https://arxiv.org/abs/2611.12956) targets DeepSeek Sparse Attention (DSA). In DSA, the indexer still scores the entire causal prefix for every query, and at long context that indexer becomes a dominant source of latency. The authors propose layer-adaptive indexing trained by group-level distillation. Earlier 2026 work on the same bottleneck reported [2x kernel speedups at 32K and 4x at 128K](https://arxiv.org/abs/2603.28458) with hierarchical indexing. [IndexCache](https://arxiv.org/abs/2603.12201) removed 75% of indexer computation on a 30B DSA model and reported a 1.82x prefill speedup. No source connects these methods to Carbon.

## What this means

Google is using its own Tensor Processing Unit (TPU) fleet and Alphabet's balance sheet to price a frontier model below Anthropic. The exposure runs through API revenue on coding workloads, which is where Anthropic earns most heavily and where tool vendors such as Cursor can switch models in one configuration change. The consensus treats the $2 rate as a temporary promotion, and that reading is correct. It misses that the standard $4/$20 rate is still half of Fable's discounted rate, so the price gap continues after November 30. The precedent is DeepSeek, which cut API prices by more than half in September 2025 when it shipped DSA in V3.2-Exp and attributed the cut to cheaper attention. The cut forced repricing discussions across Chinese providers. The timing affects capital directly: Anthropic plans to price its listing during the same week that Google undercuts its main product line. The likelier outcome is that OpenAI reprices GPT-6 Astra or ships a cheaper coding tier within weeks, and that Anthropic defends share through the lower-priced Opus 5.5 rather than by cutting Fable again. This depends on independent coding evals confirming that Carbon is at parity. The read is wrong if Carbon's usage share in Cursor and on routing platforms falls back after the December 1 price doubling.

## What to watch

- Whether Google extends the $2/$10 rate past November 30. An extension would show that Google is buying market share deliberately, not running a launch promotion.
- Independent per-task cost figures for Carbon and Fable 5.2 on agentic coding suites. Per-token prices do not show the cost of a completed task when one model uses more tokens or more agents.
- Any repricing by OpenAI of GPT-6 Astra in the next few weeks. A cut would show that the price competition has reached all three United States frontier labs.