Gemini 4 Carbon vs Claude Opus 5.5 vs GPT-6 Astra: Benchmarks and Cost
Independent Artificial Analysis numbers: Carbon 59, Opus 5.5 58, GPT-6 Astra 57 on Intelligence Index v5 — plus coding, hallucination and cost per task. Carbon has no public API yet.
29 окт. 2026 г. · API ManyThe three models sit within two points of each other on the Artificial Analysis Intelligence Index v5. The differences that matter in practice are elsewhere: what a task costs, how often each model makes things up, and whether you can use it at all.
The numbers side by side
| Artificial Analysis | Gemini 4 Carbon (high) | Claude Opus 5.5 (max) | GPT-6 Astra (max) |
|---|---|---|---|
| Intelligence Index v5 | 59 | 58 | 57 |
| Cost per Index task (standard pricing) | $7.76 | $10.04 | $6.25 |
| Hallucination rate (AA-Omniscience v1.1) | 11% | 50% | 50% |
| AA-Omniscience v1.1 score | 54 | 48 | Not stated |
| Public API | No | Yes | Yes |
Sources:Artificial Analysis: Gemini 4 Carbon is the new #1 (Oct 28, 2026) · Artificial Analysis: Announcing Intelligence Index v5 (Oct 27, 2026) · Artificial Analysis Intelligence Index leaderboard
Overall: a three-way near tie
Intelligence Index v5 weights four areas equally: knowledge work, coding, science and general (25% each). Carbon (high) leads at 59, Claude Opus 5.5 (max) is second at 58, and GPT-6 Astra (max) scored 57 in the v5 announcement. Carbon takes the top score in 4 of the 9 evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA.
Sources:Artificial Analysis: Announcing Intelligence Index v5 (Oct 27, 2026) · Artificial Analysis: Gemini 4 Carbon is the new #1 (Oct 28, 2026)
One caveat on third place: the live leaderboard summary names another Claude Opus 5.5 setting (xhigh) as third at 57, so read GPT-6 Astra’s 57 as a rounded tie for third rather than a clear #3.
Source:Artificial Analysis Intelligence Index leaderboard
Cost per task
Artificial Analysis measures what it costs to run one Index task at standard list prices. GPT-6 Astra (max) is cheapest at $6.25. Carbon costs $7.76, about 23% less than Claude Opus 5.5 (max) at $10.04 and about 1.2x GPT-6 Astra. Part of the gap is verbosity: Carbon uses 123k output tokens per task, 29% fewer than Opus 5.5 at 173k.
Source:Artificial Analysis: Gemini 4 Carbon is the new #1 (Oct 28, 2026)
Carbon’s figure uses its $4 / $20 standard price. Google’s introductory $2 / $10 rate would make it cheaper for as long as the promotion runs; Google has listed $4 / $20 as the price after the introductory period, and Artificial Analysis notes no confirmed end date.
Sources:Google: Gemini 4 Carbon announcement (Oct 28, 2026) · Artificial Analysis: Gemini 4 Carbon is the new #1 (Oct 28, 2026)
Knowledge and hallucination
This is the widest gap among the three. On AA-Omniscience v1.1, Carbon’s hallucination rate is 11% against 50% for both Claude Opus 5.5 (max) and GPT-6 Astra (max). Carbon also scores 54 overall (58% accuracy), 6 points ahead of Opus 5.5 at 48.
Source:Artificial Analysis: Gemini 4 Carbon is the new #1 (Oct 28, 2026)
Where Opus 5.5 and GPT-6 Astra led before Carbon
- HLE-Diamond: Claude Opus 5.5 (max) 68%, GPT-6 Astra (max) 64%.
- Terminal-Bench-Science: GPT-6 Astra (max) 63%, Claude Opus 5.5 (max) 59%.
- CritPt v2 hard subset: GPT-6 Astra 72% on high effort; Claude Opus 5.5 (max) 61%.
- Carbon now takes the top score on HLE-Diamond and Terminal-Bench-Science; Artificial Analysis did not state its exact percentages in the article text.
Sources:Artificial Analysis: Announcing Intelligence Index v5 (Oct 27, 2026) · Artificial Analysis: Gemini 4 Carbon is the new #1 (Oct 28, 2026)
Which to pick, by use case
- You need it in production this week: Claude Opus 5.5 or GPT-6 Astra. Carbon cannot be called yet.
- Lowest cost per task among the three: GPT-6 Astra ($6.25 per Index task).
- Fewest made-up answers on knowledge questions: Carbon (11% hallucination), once it is available. Of the two available models, neither stands out: both measured 50%.
- Research-style physics and science tasks today: GPT-6 Astra led CritPt and Terminal-Bench-Science among the models measured before Carbon.
- Coding agents: Claude Opus 5.5. Claude Code with Opus 5.5 (max) scores 62 on the Artificial Analysis Coding Agent Index, ahead of Carbon at 59; the head-to-head below goes into detail.
Sources:Artificial Analysis: Gemini 4 Carbon is the new #1 (Oct 28, 2026) · Artificial Analysis: Announcing Intelligence Index v5 (Oct 27, 2026)
Related
- Gemini 4 Carbon: what it is and how to get accessSpecs, prices, the access rollout and Google’s own benchmark results.
- Gemini 4 Carbon vs Claude Opus 5.5A closer look at coding agents, long tasks and security work.
FAQ
Which model is #1 on the Artificial Analysis Intelligence Index?
Gemini 4 Carbon (high) at 59, ahead of Claude Opus 5.5 (max) at 58. GPT-6 Astra (max) scored 57 in the v5 announcement.
Which of the three is cheapest to run?
GPT-6 Astra (max), at $6.25 per Intelligence Index task at standard pricing, versus $7.76 for Carbon and $10.04 for Opus 5.5 (max).
Can I call Gemini 4 Carbon today?
No. It has no public API and no published model ID. Access is limited to selected cyber defenders and Fairwind Program partners for now.
Which of these can I use on API Many?
Claude Opus 5.5 and GPT-6 Astra, plus Gemini 4 Argon. Gemini 4 Carbon is not available on API Many.
Sources
- Artificial Analysis: Gemini 4 Carbon is the new #1 (Oct 28, 2026) — Carbon index score and evaluation leads, cost per task, output tokens, AA-Omniscience and hallucination rates.
- Artificial Analysis: Announcing Intelligence Index v5 (Oct 27, 2026) — Index v5 weights; Opus 5.5, GPT-6 Astra and Argon scores; HLE-Diamond, Terminal-Bench-Science and CritPt results.
- Artificial Analysis Intelligence Index leaderboard — Live leaderboard order and the “No publicly available API” label for Carbon.
- Google: Gemini 4 Carbon announcement (Oct 28, 2026) — Access plan and introductory / standard prices.
- Google DeepMind: Fairwind Program — Fairwind Program access.
- Google AI for Developers: Gemini API models — Public Gemini API models; Carbon not listed as of October 29, 2026.
Все модели из статьи доступны в API Many. Оплата по факту от $10.