Claude Fable 5.2 vs Gemini 4 Carbon: Tough Time for Google Too
Google recently announced its Gemini 4 Argon and Carbon models back to back with significant improvements, but it is a hard time for them again. Google took some time to announce these models, and Anthropic has quickly taken the model crown with its latest Fable 5.2 model. Here is how the latest Claude Fable 5.2 compares with Gemini 4 Carbon.
Benchmarks
| Benchmark | Claude Fable 5.2 | Gemini 4 Carbon | Difference |
|---|---|---|---|
| FrontierCode 2.0 | 49.1%Xhigh effort | 41.1%Medium effort | Fable +8.0 points |
| FrontierSWE v2 | 69.9% | 61.2% | Fable +8.7 points |
| AutomationBench | 51.4% | 53.9% | Carbon +2.5 points |
| Terminal-Bench-Science 0.1 | 74.6% | 64.8% | Fable +9.8 points |
| OSWorld 2.1 | 89.3% | 81.5% | Fable +7.8 points |
| AA-Briefcase v1.1 rubric | 66.1% | 61.8% | Fable +4.3 points |
| AA-Briefcase v1.1 Elo | 1785 Elo | 1611 Elo | Fable +174 Elo |
| Chartography | 73.4% | 72.5% | Fable +0.9 points |
Scores from Anthropic. Fable 5.2 leads on 7 of 8, and Gemini 4 Carbon wins AutomationBench.
Artificial Analysis benchmarks
| Benchmark | Fable 5.2, Max effort | Gemini 4 Carbon, High effort | Difference |
|---|---|---|---|
| Intelligence Index v5.0.1 | 65 | 60 | Fable +5 |
| AA-Briefcase v1.1 | 1785 Elo | 1611 Elo | Fable +174 Elo |
| AutomationBench-AA | 76.4% | 77.8% | Carbon +1.4 points |
| Terminal-Bench 4.0 | 59.6% | 65.2% | Carbon +5.6 points |
| FrontierCode 2.0 | 48.8% | 40.6% | Fable +8.2 points |
| AA-Omniscience v1.1 | 67 | 54 | Fable +13 |
| GDP.pdf | 24.2% | 28.4% | Carbon +4.2 points |
| HLE-Diamond | 80.2% | 69.1% | Fable +11.1 points |
| CritPt v2 | 76.5% | 69.0% | Fable +7.5 points |
| Terminal-Bench-Science 0.1 | 73.3% | 64.8% | Fable +8.5 points |
Fable 5.2 wins 7 of these 10. Gemini 4 Carbon wins AutomationBench-AA, Terminal-Bench 4.0 and GDP.pdf.
Cost and speed
| Metric | Fable 5.2, Max effort | Gemini 4 Carbon, High effort | Difference |
|---|---|---|---|
| Cost per Intelligence Index task | $26.54 | $7.76 | Carbon about 71% cheaper |
| Cost to run the full Index | $22,191 | $8,240 | Carbon about 63% cheaper |
| Output tokens per task | 274K | 123K | Carbon about 55% fewer |
| Reasoning tokens per task | 193K | 72.2K | Carbon about 63% fewer |
| Output tokens for the full Index | 275M | 205M | Carbon about 25% fewer |
| Output speed | 103 tokens/s | 75 tokens/s | Fable about 37% faster |
| Time to first token | 231.55s | 69.21s | Carbon about 70% sooner |
| Average time per task | 44.1 minutes | 27.2 minutes | Carbon about 38% faster |
Intelligence at each effort level
| Model and effort | Intelligence Index | Cost per task |
|---|---|---|
| Fable 5.2, Max | 65 | $26.54 |
| Fable 5.2, Xhigh | 62 | $14.83 |
| Fable 5.2, High | 62 | $9.08 |
| Gemini 4 Carbon, High | 60 | $7.76 |
| Gemini 4 Carbon, Medium | 58 | $6.59 |
| Fable 5.2, Medium | 56 | $3.80 |
| Gemini 4 Carbon, Low | 54 | $7.19 |
| Fable 5.2, Low | 54 | $3.10 |
At High effort, Fable 5.2 still scores 2 points above Carbon at High for about $1.30 more per task. At Low, both score 54, and Fable 5.2 costs less than half as much.
Pricing and specs
| Spec | Claude Fable 5.2 | Gemini 4 Carbon | Difference |
|---|---|---|---|
| Input | $8 | $2Introductory, then $4 | Carbon 75% cheaper for now |
| Output | $40 | $10Introductory, then $20 | Carbon 75% cheaper for now |
| Cached input | $0.20 | $0.10Introductory, then $0.20 | Carbon 50% cheaper for now |
| Context window | 1M tokens | 1M tokens | Same |
| Max output | 128K tokens | Not published | - |
| Reasoning | Adaptive, always onLow to Max effort | Low, Medium and High | Both |
| Input types | Text and images | Text, images, speech and video | Carbon also takes audio and video |
| Output | Text | Text | Same |
| Fast mode | Yes2x price, up to 2.5x faster | Not announced | Fable |
| Public API | Yesclaude-fable-5-2 | Early accessCyber defenders first | Fable |
| Released | November 9, 2026 | October 28, 2026 | Fable 12 days newer |