Get started

GPT-5.5 (batch) against Claude Fable 5.2.

Two listings from the same committed catalog, side by side: the fields OpenRouter publishes, the boards two third parties scored, and a cost sheet that is arithmetic over the listed rates.

retrieved sha256 · 0bb53a77713f

Compare/GPT-5.5 (batch)vsClaude Fable 5.2USD per 1M tokens · one scale
Listed
1,050,000 tokens
Context
1,000,000 tokens
128,000 tokens
Max output
128,000 tokens
file, image, text → text
Modalities
text, image, file → text
2025-12-01
Knowledge cutoff
not stated
45.2 tok/s
Throughput
84.9 tok/s
3.87s
First token
4.71s
1,722 requests · 5 endpoints
Timed over
12,305 requests · 5 endpoints

Listed price

USD per million tokens · one linear scale
GPT-5.5 (batch)OpenAI · listed 24 Apr 2026
Input per M$2.50
Cached input per M$0.25
Output per M$15.00
Blended, 3 in to 1 out$5.625
Claude Fable 5.2Anthropic · listed 9 Nov 2026
Input per M$8.00
Cached input per M$0.20
Output per M$40.00
Blended, 3 in to 1 out$16.00

Per million tokens, as OpenRouter listed them when the catalog was read: its routed rates, not a provider's own page. One linear scale across every row, so a bar twice as long is twice the price. The blended row is Vitna's arithmetic, three input tokens to one output.

Every board either stands on

27 boards
not scored
Intelligence Index
64.6 · 01 of 58
74.9 · 12 of 118
Coding Index
not scored
36.4 · 25 of 94
Agentic Index
not scored
1247 · 44 of 126
Websites · Models
1350 · 03 of 126
1248 · 43 of 120
Code · Models
1374 · 03 of 120
1281 · 22 of 120
Game development · Models
1401 · 05 of 120
1248 · 43 of 116
UI components · Models
1378 · 03 of 116
1175 · 56 of 113
3D scenes · Models
1498 · 01 of 113
1258 · 30 of 110
Data visualisation · Models
not ranked
1225 · 22 of 83
SVG · Models
not ranked
1235 · 21 of 67
ASCII art · Models
not ranked
1134 · 35 of 49
Mobile apps · Agents
1318 · 02 of 49
1097 · 38 of 47
Web apps · Agents
not ranked
1063 · 32 of 44
Full stack · Agents
1397 · 01 of 44
1153 · 26 of 41
Android native · Agents
not ranked
not ranked
pdf · Agents
1351 · 01 of 36
1142 · 21 of 34
Godot game development · Agents
not ranked
not ranked
spreadsheet · Agents
1429 · 01 of 32
1152 · 20 of 29
python-pptx slides · Agents
not ranked
not ranked
codetovideo · Agents
1445 · 01 of 26
1124 · 15 of 25
Agentic game development · Agents
not ranked
964 · 24 of 25
HTML slides · Agents
not ranked
1084 · 08 of 9
Agentic HTML slides · Agents
not ranked
1150 · 06 of 9
Agentic slides · Agents
not ranked
1077 · 08 of 9
Agentic slides, HTML · Agents
not ranked
1155 · 06 of 9
Agentic slides, python-pptx · Agents
not ranked
1157 · 06 of 8
PPTX slides · Agents
not ranked

Score and position on each board, from the board's own source as OpenRouter carries it. The heavier figure leads that board; a missing score is a fact about coverage, not a zero.

Cost sheet

800 in · 200 out per call

Arithmetic over the listed rates and nothing else: tokens in over one million times the input rate, plus tokens out over one million times the output rate. No cache, no retries and no failure rate assumed; the Decision Lab is where cost per successful unit lives.

ModelPer callPer 1,000 callsPer 100,000 calls
GPT-5.5 (batch)$0.005000$5.00$500
Claude Fable 5.2$0.0144$14.40$1,440

GPT-5.5 (batch) costs 65% less per call at these token counts.

Other pairs with a page
For one of your workloads

The market on these two is here. The right one for your work is not.

What this page cannot say is which is right for a particular piece of your production work, because that depends on what the workload requires and what your own evaluations measured, neither of which is public. The register weighs a candidate against a workload's own thresholds and shows what the engine would decide before anything is filed.

Sources

Source: OpenRouter (openrouter.ai/models), as of 17 Nov 16:29 UTC. sha256 0bb53a77713f over 853,270 bytes.
Catalog: OpenRouter's public model list.
Intelligence, coding and agentic indexes: Artificial Analysis, as published by OpenRouter.
Elo, win rate and rank: Design Arena, as published by OpenRouter.
Vitna ran none of these evaluations and adjusted none of these figures.