vs
4:0
overall capability wins (2 ties across 6 scored). Claude Fable 5.1 leads overall.
- Reasoning2:0
- Coding2:0
Verdict
Claude Fable 5.1 leads coding (widest gap: +7.6 on TermBench).
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Claude Fable 5.1
- $22.50
- Claude Mythos 5
- $22.50
Price per 1M tokens
Full comparison
Organization
Anthropic
Anthropic
Family
Claude Fable / Mythos
Claude Fable / Mythos
License
Proprietary
Proprietary
Open weights
No
No
Release
Sep 1, 2026
Jun 9, 2026
Knowledge cutoff
-
-
API / provider
Anthropic
Anthropic
Modalities
text, image → text
text, image → text
Specs
Context window
1M
1M
Max output
128K
128K
Parameters
—
—
Pricing
Input $/1M
$10
$10
Output $/1M
$50
$50
Blended $/1M (3∶1)
$20
$20
Speed
tok/s
-
55
TTFT (s)
-
0.75
Reasoning
MMLU-Pro
—
—
GPQA Diamond
93.7
92.6
Humanity's Last Exam
59.0
32.5
AIME 2025
—
97.0
MATH-500
—
—
Humanity's Last Exam (with tools)
65.6
—
AA-Omniscience Accuracy
67.2
—
AA-LCR v1.1
83.0
—
CritPt
29.7
—
MMMU-Pro
—
—
IFBench
—
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
86.3
SWE-bench Pro
—
80.0
SWE-bench Multilingual
—
—
LiveCodeBench
—
88.1
Terminal-Bench 2.1
91.4
83.8
Aider Polyglot
—
—
Terminal-Bench 3
—
34.1
BigCodeBench
—
—
SciCode
63.0
—
CursorBench
73.4
70.5
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
67.4
—
WebDev Arena
1,758
—
Terminal-Bench 4.0
57.9
—
CursorBench 4.0
51.8
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,499
—
| Metric | Claude Fable 5.1 | Claude Mythos 5 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Anthropic | Anthropic | — |
| Family | Claude Fable / Mythos | Claude Fable / Mythos | — |
| License | Proprietary | Proprietary | — |
| Open weights | No | No | — |
| Release | Sep 1, 2026 | Jun 9, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Anthropic | Anthropic | — |
| Modalities | text, image → text | text, image → text | — |
| Specs | |||
| Context window | 1M | 1M | tie |
| Max output | 128K | 128K | tie |
| Parameters | — | — | — |
| Pricing | |||
| Input $/1M | $10 | $10 | tie |
| Output $/1M | $50 | $50 | tie |
| Blended $/1M (3∶1) | $20 | $20 | tie |
| Speed | |||
| tok/s | - | 55 | — |
| TTFT (s) | - | 0.75 | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 93.7 | 92.6 | A +1.1 pts |
| Humanity's Last Exam | 59.0 | 32.5 | A +26.5 pts |
| AIME 2025 | — | 97.0 | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | 65.6 | — | — |
| AA-Omniscience Accuracy | 67.2 | — | — |
| AA-LCR v1.1 | 83.0 | — | — |
| CritPt | 29.7 | — | — |
| MMMU-Pro | — | — | — |
| IFBench | — | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | 86.3 | — |
| SWE-bench Pro | — | 80.0 | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | 88.1 | — |
| Terminal-Bench 2.1 | 91.4 | 83.8 | A +7.6 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | 34.1 | — |
| BigCodeBench | — | — | — |
| SciCode | 63.0 | — | — |
| CursorBench | 73.4 | 70.5 | A +2.9 pts |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | 67.4 | — | — |
| WebDev Arena | 1,758 | — | — |
| Terminal-Bench 4.0 | 57.9 | — | — |
| CursorBench 4.0 | 51.8 | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,499 | — | — |
Claude Fable 5.1
- Context
- 1M / 128K out
- Parameters
- —
- Price
- $10 / $50
- Speed
- —
- Modalities
- text, image → text
- License
- Proprietary
Anthropic's September 2026 successor to Fable 5 - same input/output pricing with cache reads cut to a quarter of the price, and gains concentrated in long-horizon agentic coding, knowledge work, research, and computer use.
Claude Mythos 5
- Context
- 1M / 128K out
- Parameters
- —
- Price
- $10 / $50
- Speed
- 55 tok/s · 0.75s TTFT
- Modalities
- text, image → text
- License
- Proprietary
Limited-availability Mythos-class model (Project Glasswing) - Fable's unrestricted sibling for trusted partners.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.