Compare / head-to-head
Claude Fable 5vs
Inkling-Small
Claude Fable 5 leads 13 of 14 shared benchmarks. Inkling-Small is 40x cheaper per token. Both offer a 1M-token context window.
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
13 – 1
Claude Fable 5 leads
Cheaper per token
Inkling-Small
40x cheaper, input + output
Larger context
Tie
both 1M tokens
Providers
2 providers
Anthropic · Thinking Machines Lab
Claude Fable 5
Anthropic · released 2026-06-09
textvision
Inkling-Small
Thinking Machines Lab · released 2026-07-30
textvisionaudio
Quality
Benchmark matrix
14 shared · 30 only Claude Fable 5 · 14 only Inkling-Small| Benchmark | Claude Fable 5 | Inkling-Small | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 14 | ||||
| SWE-bench Verified | 95% ↗ | 80.2% ↗ | +14.8 pt | Claude Fable 5 |
| GPQA Diamond | 85.9% ↗epoch run | 88.5% ↗epoch run | -2.6 pt | Inkling-Small |
| LMArena Elo | 1504.3 ↗ | 1405.2 ↗ | +99.1 | Claude Fable 5 |
| Chess Puzzles (Epoch AI run) | 41% ↗epoch run | 18% ↗epoch run | +23 pt | Claude Fable 5 |
| FrontierMath Tier 4 v2 (Epoch AI run) | 90.2% ↗epoch run | 17.1% ↗epoch run | +73.1 pt | Claude Fable 5 |
| FrontierMath Tiers 1-3 v2 (Epoch AI run) | 87% ↗epoch run | 46.3% ↗epoch run | +40.7 pt | Claude Fable 5 |
| LMArena Agent (LMArena) | 0.0821 ↗ | -0.1033 ↗ | +0.2 | Claude Fable 5 |
| LMArena Vision (LMArena) | 1308.8 ↗ | 1206.3 ↗ | +102.5 | Claude Fable 5 |
| LMArena WebDev (LMArena) | 1625.3 ↗ | 1408.2 ↗ | +217.1 | Claude Fable 5 |
| Mystery Game Puzzles (Epoch AI run) | 52% ↗epoch run | 6% ↗epoch run | +46 pt | Claude Fable 5 |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 100% ↗epoch run | 90% ↗epoch run | +10 pt | Claude Fable 5 |
| SAGE (Vals AI) | 51.9% ↗ | 32.7% ↗ | +19.2 pt | Claude Fable 5 |
| SimpleQA Verified | 70.7% ↗epoch run | 20.6% ↗ | +50.1 pt | Claude Fable 5 |
| Terminal-Bench 4.0 (Vals AI) | 41.4% ↗ | 0% ↗ | +41.4 pt | Claude Fable 5 |
| Only Claude Fable 5 reports · 30 | ||||
| APEX-Agents (Mercor) | 63.6% ↗ | not reported | — | — |
| AutomationBench | 17.4% ↗ | not reported | — | — |
| Blueprint-Bench 2 | 38.6% ↗ | not reported | — | — |
| CursorBench | 72.9% ↗ | not reported | — | — |
| DeepSWE v1.1 (Datacurve) | 69.9% ↗ | not reported | — | — |
| EBR-bench (Epoch AI run) | 39.5% ↗epoch run | not reported | — | — |
| FrontierCode (Diamond) | 29.3% ↗ | not reported | — | — |
| FrontierSWE V2 (Proximal Labs) | 47% ↗ | not reported | — | — |
| Furniture Assembly (Epoch AI run) | 35.8% ↗epoch run | not reported | — | — |
| GDP.pdf | 29.8% ↗ | not reported | — | — |
| GSO Opt@1 (GSO) | 76.5% ↗ | not reported | — | — |
| Legal Agent Benchmark (Harvey's Held-Out Set) | 13.3% ↗ | not reported | — | — |
| LiveBench Agentic Coding (LiveBench) | 62.2% ↗ | not reported | — | — |
| LiveBench Coding (LiveBench) | 86% ↗ | not reported | — | — |
| LiveBench Data Analysis (LiveBench) | 80.5% ↗ | not reported | — | — |
| LiveBench Instruction Following (LiveBench) | 75.8% ↗ | not reported | — | — |
| LiveBench Language (LiveBench) | 90.7% ↗ | not reported | — | — |
| LiveBench Mathematics (LiveBench) | 96% ↗ | not reported | — | — |
| LiveBench Reasoning (LiveBench) | 89.7% ↗ | not reported | — | — |
| MirrorCode (Epoch AI run) | 63.9% ↗epoch run | not reported | — | — |
| OfficeQA Pro | 57.9% ↗ | not reported | — | — |
| OSWorld-Verified | 85% ↗ | not reported | — | — |
| OSWorld-Verified (XLANG) | 86% ↗ | not reported | — | — |
| SimpleBench (SimpleBench) | 81.9% ↗ | not reported | — | — |
| SWE-Bench Pro | 80% ↗ | not reported | — | — |
| SWE-rebench 2026-05-15 to 2026-07-01 (Nebius) | 64.5% ↗ | not reported | — | — |
| tau2-bench Banking Knowledge (Sierra) | 39.7% ↗ | not reported | — | — |
| Terminal-Bench 2.1 | 84.3% ↗ | not reported | — | — |
| Vending-Bench 2 (Andon Labs) | 5680.26 ↗ | not reported | — | — |
| WeirdML (Håvard Tveit Ihle) | 91.9% ↗ | not reported | — | — |
| Only Inkling-Small reports · 14 | ||||
| AIME 2026 | not reported | 95.5% ↗ | — | — |
| ARC-AGI-1 | not reported | 84% ↗ | — | — |
| ARC-AGI-2 | not reported | 40.1% ↗ | — | — |
| BrowseComp (with context management) | not reported | 77.4% ↗ | — | — |
| CharXiv RQ | not reported | 77.4% ↗ | — | — |
| CharXiv RQ (with python) | not reported | 81.3% ↗ | — | — |
| HMMT Feb 2026 | not reported | 90.2% ↗ | — | — |
| IFBench | not reported | 82.2% ↗ | — | — |
| MCP Atlas (all) | not reported | 79.2% ↗ | — | — |
| MMAU | not reported | 77% ↗ | — | — |
| SWE-bench Pro (public) | not reported | 55.9% ↗ | — | — |
| Terminal-Bench 2.1 (best harness) | not reported | 64.7% ↗ | — | — |
| Toolathlon Verified | not reported | 54.4% ↗ | — | — |
| Toolathlon-Verified (HKUST) | not reported | 54.4% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is Claude Fable 5 minus Inkling-Small in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | Claude Fable 5 | Inkling-Small | Edge |
|---|---|---|---|
| Context window | 1M tokens | 1M tokens | Tie |
| Max output | 128K tokens | — | — |
| Input price / 1M | $10 ↗ | $0.3 ↗ | Inkling-Small |
| Output price / 1M | $50 ↗ | $1.2 ↗ | Inkling-Small |
| Cached input / 1M | $1 ↗ | $0.06 ↗ | Inkling-Small |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $60 | $1.5 | Inkling-Small |
| Modalities | text · vision | text · vision · audio | Inkling-Small |
| Released | 2026-06-09 | 2026-07-30 | — |
| Cited benchmark scores | 44 | 29 | — |
Reliability
Provider status
Live from /statusMore matchups
Claude Fable 5 vs …
Models sharing the most benchmarksMore matchups
Inkling-Small vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run Claude Fable 5 and Inkling-Small on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.