Claude Fable 5.1 Tops October LiveBench at 83.4, but GPT-6.1 Sol Delivers 98% of That at 12% of the Cost
Claude Fable 5.1 in Max Effort mode leads the October LiveBench leaderboard at 83.4 overall, edging Claude 5.5 Opus Thinking Max Effort (83.2) and Claude Fable 5 Max Effort (83.0). The top three positions are all Anthropic. GPT-6 Astra Max Effort follows at 82.2 and GPT-6.1 Sol Max Effort at 81.6, alongside Meta’s Muse Spark 1.3 in xHigh Effort, also at 81.6.
The number that matters most is not in the overall column. It is in cost per successful task: Fable 5.1 Max Effort costs $1.212. GPT-6.1 Sol Max Effort costs $0.142. That is an 8.5x gap for 1.8 percentage points of overall performance.
Where OpenAI’s Flagship Falls Short
LiveBench’s Agentic Coding category is the single most diagnostic column in this snapshot. It measures autonomous software work, not pattern completion. The spread across the top five is 17 points wide:
| Model | Overall | Agentic Coding | Cost/task |
|---|---|---|---|
| Claude Fable 5.1 Max Effort | 83.4 | 66.1 | $1.212 |
| Claude 5.5 Opus Thinking Max Effort | 83.2 | 71.7 | $0.799 |
| Claude Fable 5 Max Effort | 83.0 | 62.2 | $1.439 |
| GPT-6 Astra Max Effort | 82.2 | 57.3 | $0.736 |
| GPT-6.1 Sol Max Effort | 81.6 | 54.5 | $0.142 |
GPT-6 Astra leads or ties the field in reasoning (92.7), mathematics (96.8), and data analysis (83.0). The agentic coding gap is model-specific, not category-wide. OpenAI’s flagship produces better raw reasoning than any Anthropic model in this snapshot, and that reasoning does not convert to autonomous coding performance at the same rate.
Opus 5.5 Thinking’s 71.7 on agentic coding is the highest score in this leaderboard slice. Fable 5.1 sits 5.6 points behind it on that category despite leading overall, because its language (89.5 vs 86.3) and instruction following (73.0 vs 65.7) scores tip the composite.
The Fable 5.1 Lead Is Real but Narrow
Fable 5.1’s overall win over Opus 5.5 Thinking (83.4 vs 83.2) rests on two categories: language and instruction following. Opus 5.5 Thinking beats or ties Fable 5.1 in reasoning, coding, agentic coding, and mathematics. Fable 5.1 is not a better model in every dimension — it is a more balanced model. Whether that balance matters depends on the task.
Claude Fable 5, the predecessor, sits at 83.0 overall but posts the highest instruction following score (75.8) and the best language score (90.7) of any model in the top five. It is also the most expensive at $1.439 per task. The Fable 5 to Fable 5.1 transition improved agentic coding (62.2 to 66.1) and reasoning at modest cost to instruction following (75.8 to 73.0).
Meta’s Muse Spark 1.3 Enters the Conversation
Muse Spark 1.3 in xHigh Effort ties GPT-6.1 Sol at 81.6 overall — placing a Meta model in the top five on LiveBench’s composite leaderboard. This is the second distinct Muse model line to appear at frontier-adjacent performance: Muse Wonder is Artificial Analysis’s mid-20s range on intelligence index, while Muse Spark 1.3 is more cost-efficient and appears to index differently. The October LiveBench snapshot does not break out Muse Spark 1.3’s per-category scores in the top-line view.
The Practical Cost Calculation
For teams running agent workloads at scale, the cost-per-task metric changes the procurement math significantly. At Opus 5.5 Thinking’s $0.799 per task and Sol’s $0.142, the question is not whether Fable 5.1’s 83.4 is better than Sol’s 81.6 — it is whether the 1.8-point difference justifies an 8.5x cost premium. For agentic coding specifically, Sol’s 54.5 versus Opus 5.5 Thinking’s 71.7 represents a larger quality gap than the overall score suggests, and Opus 5.5 Thinking delivers that at $0.799, not $1.212.
The October snapshot makes the case for Opus 5.5 Thinking as the value leader at the frontier: best agentic coding score in the group, overall score 0.2 points below Fable 5.1, and 34% cheaper than Fable 5.1.