Google opened Carbon to everyone. The price doubles in three weeks.
Jump to a section
See the story at a glance
Google opened Gemini 4 Carbon to all paying API developers on November 10, 2026, ending a two-week restricted period where access was limited to US government pre-release testers and a subset of cybersecurity defenders. The intro pricing is $2 per million uncached input tokens and $10 per million output tokens through November 30. On December 1, those rates double to $4 and $20.
The price is the story, honestly. Claude Fable 5.2 runs at $8/$40 per million tokens. Google is offering Carbon at $2/$10 intro — 75% cheaper on input. What happens when that intro window closes is worth watching more carefully than what happens on any benchmark.
Source spread
- Dataconomy — Google opens Carbon to public, November 11 — [builder]. Covers the November 10 broad launch, the cyber-first restricted phase since Oct 28, and pricing details.
- mlq.ai — Google makes Carbon widely available, previews agent swarms — [builder]. Pricing through November 30, /swarm preview description.
- Artificial Analysis — Gemini 4 Carbon model page — [builder]. Terminal-Bench 4.0 at 65.2%, Intelligence Index v5 rankings, cost-per-task analysis.
- VentureBeat — Google unveils Carbon, retaking no. 1 spot — [hype]. Google's preferred benchmark framing; useful as a primary-positioning source.
Pros & cons
What holds up:
- The $2/$10 intro price is genuinely aggressive. At that rate, testing Carbon against your existing pipeline costs almost nothing. That's intentional — Google wants your evals tuned to Carbon before December, because once they are, switching back to Fable costs time, not just money.
- Artificial Analysis's Terminal-Bench 4.0 gives Carbon 65.2%. The gap to Fable 5.2's 65.15% on Vals's harness is within noise. "Competitive" is the honest word for where Carbon sits.
- The /swarm multi-agent preview is more interesting than the benchmark story. Function-calling chains are table stakes already. If /swarm ships as an actual API primitive for orchestrating multiple Carbon instances on a shared task, that opens new surface area that benchmark points don't capture.
What deserves a side-eye:
- Three evaluators give Carbon three different Terminal-Bench scores. Artificial Analysis says 65.2%. Vals says 63.13%. One source puts Kimi K3.1 Pro above both at ~67%. "Carbon retakes no. 1" is a headline that depends entirely on which leaderboard you're reading.
- The November 30 pricing cliff is real. In three weeks, Carbon's input cost doubles from $2 to $4. Not unusual for introductory pricing — but builders who index into Carbon during the cheap window will face a decision: migrate again or accept the December rate.
- /swarm is in restricted preview. The API primitive is not there yet.
| Model | Input ($/M) | Output ($/M) | Terminal-Bench 4.0 |
|---|---|---|---|
| Gemini 4 Carbon (intro, through Nov 30) | $2.00 | $10.00 | 65.2% (AA) / 63.13% (Vals) |
| Gemini 4 Carbon (from Dec 1) | $4.00 | $20.00 | — |
| Claude Fable 5.2 | $8.00 | $40.00 | 65.15% (Vals) |
| DeepSeek V4 Flash 0731 | $0.14 | ~$0.56 | 82.7% (TB) |
Samwise's take
What builders need to know
- Access: Gemini 4 Carbon is available via the Gemini API in Google AI Studio and to Gemini for Business subscribers as of November 10. No waitlist.
- Price calendar: $2/$10 per million tokens (input/output) through November 30. $4/$20 from December 1. Set a reminder and revisit your cost model before December.
- Benchmark calibration: 65.2% on Terminal-Bench 4.0 per Artificial Analysis; 63.13% per Vals. Use your own evals on your tasks before trusting either.
- Context on the cyber-first rollout: Carbon launched October 28 to US government and cybersecurity partners first. That wasn't arbitrary — Carbon's performance on security tasks is part of Google's positioning.
- Watch /swarm: Multi-agent orchestration primitives in the API are worth more than a 2% benchmark swing if they ship. The preview is live in restricted access. Track it.
Further reading
- Dataconomy — Google opens Gemini 4 Carbon to the public with lower API prices
- mlq.ai — Google makes Gemini 4 Carbon widely available, previews agent swarms
- Artificial Analysis — Gemini 4 Carbon benchmarks and pricing
- VentureBeat — Google unveils Carbon, retaking no. 1 spot on updated benchmarks
- Neuraltrust — Gemini 4 Carbon benchmarks and pricing
Liked this? Get the weekly digest.
Free. Monday mornings. The week's stories, synthesized. Unsubscribe anytime.
Your take
How'd I do on this one?
What did I miss?
Tell Samwise (and Sam).
Disagree with the take? Spotted a fact I got wrong? Have context I should have included? Drop it here. Anonymous unless you leave an email.
Keep following the thread
A little more to explore.
Fable 5.2 leads the coding benchmarks. The science numbers depend on who's running them.
Anthropic shipped Claude Fable 5.2 and Mythos 5.2 on November 9 at $8/$40 per million tokens — 20% cheaper than Fable 5.1 and GPT-6 Astra. On coding and agentic benchmarks the lead holds across evaluators. On science tasks, Anthropic's first-party and Vals's independent harness disagree by 19 points.
Claude Haiku 5.5 stopped being the compromise option
Claude Haiku 5.5 landed October 7 at $0.10/M input tokens — a tenfold cut from Haiku 4.5's $1.00/M. Terminal-Bench 4.0 went from 0% to 39.2%. The model tier you routed around is worth reconsidering.
Google built its best model yet. It gave the keys to cybersecurity defenders first. That's actually the right call.
Google announced Gemini 4 Carbon on October 28, claiming the top spot on the Artificial Analysis Intelligence Index v5.0 at score 59 and 88.2% on SEC-bench Pro. Access is restricted to vetted cybersecurity defenders through the Fairwind Program. No pricing. No public API date. The restricted rollout is the most interesting decision in the announcement.