Skip to main content
ModelChoice

ModelChoice

Choose an AI model based on evidence, not a launch post.

Compare benchmark scores, response times, and API prices. Check where each number comes from and see which measurements are missing.

1,177 models and 534 benchmarks from 21 tracked sources · generated 2026-11-22

  1. Claude Fable 5.2Anthropic · rank 1
    Overall score
    93.3
    Time to first token
    245.87 s
    Price per 1M output tokens
    $40.00
  2. Gemini 4 NeonGoogle · rank 2
    Overall score
    91.3
    Time to first token
    56.51 s
    Price per 1M output tokens
    $15.00
  3. GPT-6.2 AstraOpenAI · rank 3
    Overall score
    90.8
    Time to first token
    Unavailable
    Price per 1M output tokens
    $40.00

Model releases and retirements

Data refreshed . Release and retirement dates come straight from the sources; “From” marks a date the provider gives only as the earliest possible.

New releases

Catalog
  • claude-sonnet-4-5-20250929Anthropic
  • gemini-2.5-flashGoogle
  • gemini-2.5-proGoogle
  • chatgpt-image-latestOpenAI

Leaderboards

Model rankings: overall and by task

Each ranking includes only models with a published score for that category.

All leaderboards

Top 15 models

Benchmark score vs. token cost

Each point is a model with an overall benchmark score and published input and output prices. Higher means a higher score; further left means a lower price per 1M tokens (frontier models are spaced evenly along the axis), assuming a mix of 75% input / 25% output tokens. The line joins the models that no other model shown beats on both score and price. The overall score summarises published benchmarks; it does not measure every aspect of quality.

20406080$0.034$0.055$0.088$0.13$0.15$0.18$0.2$0.26$0.52$2$4$6$16Blended $ / 1M tokens (75% input / 25% output)Overall score

Top three by capability

Explore models by access and speed

This preview covers 477 models: those with an overall score, a BenchLM verified score, a token price, or a throughput measurement. The workbench lists 1,010 models with any published data, out of 1,177 in the full catalog.

Explore all 1,010 models with published data

Rule: 477 of 1,177 catalog models — those with at least one of an overall score, a BenchLM verified score, a token price, or a throughput measurement.

Evidence key: Observed

Time to first token (TTFT) is the wait before an answer starts, measured to the first chunk of the answer (Artificial Analysis, via BenchLM). For reasoning models it includes thinking time, so values can run to tens or hundreds of seconds.

All models
ModelAccessScore$/1M in$/1M outTTFTThroughput
Claude Fable 5.2AnthropicProprietary
93.3$8.00$40.00245.87 s109 tok/s
Gemini 4 NeonGoogleProprietary
91.3$3.00$15.0056.51 s110 tok/s
GPT-6.2 AstraOpenAIProprietary
90.8$8.00$40.00Unavailable51 tok/s
Gemini 4 CarbonGoogleProprietary
88.3$2.00$10.0071.28 s55 tok/s
GPT-6 AstraOpenAIProprietary
83.6$10.00$50.00204.16 s42 tok/s
Claude Opus 5.5AnthropicProprietary
82.8$4.00$20.00352.74 s113 tok/s

Compare top-ranked models

Choose two to four models to compare. Select a model name to see its scores and sources.

Full leaderboard

Claude Fable 5.2

Anthropic · rank 1

ProprietaryCodingReasoning
Context1M tokens
Input / 1M$8.00
Output / 1M$40.00

Runtime

Overall93.3
Verified (BenchLM)87
TTFT245.87 s
Throughput109 tok/s

Capability

AgenticUnavailable
Coding88
Knowledge94
Reasoning86.8
Multimodal & Grounded89.3
Instruction FollowingUnavailable
Math93.8

Gemini 4 Neon

Google · rank 2

Proprietary
Context1M tokens
Input / 1M$3.00
Output / 1M$15.00

Runtime

Overall91.3
Verified (BenchLM)77
TTFT56.51 s
Throughput110 tok/s

Capability

AgenticUnavailable
CodingUnavailable
Knowledge83.1
ReasoningUnavailable
Multimodal & GroundedUnavailable
Instruction FollowingUnavailable
Math92.5

GPT-6.2 Astra

OpenAI · rank 3

ProprietaryReasoning
Context1.05M tokens
Input / 1M$8.00
Output / 1M$40.00

Runtime

Overall90.8
Verified (BenchLM)82
TTFTUnavailable
Throughput51 tok/s

Capability

AgenticUnavailable
CodingUnavailable
Knowledge84.9
Reasoning91.4
Multimodal & Grounded87.8
Instruction FollowingUnavailable
Math93.8

Gemini 4 Carbon

Google · rank 4

ProprietaryReasoning
Context1M tokens
Input / 1M$2.00
Output / 1M$10.00

Runtime

Overall88.3
Verified (BenchLM)80
TTFT71.28 s
Throughput55 tok/s

Capability

AgenticUnavailable
CodingUnavailable
Knowledge80.6
Reasoning88.2
Multimodal & GroundedUnavailable
Instruction FollowingUnavailable
Math91.6

GPT-6 Astra

OpenAI · rank 5

ProprietaryCodingAgentsReasoning
Context1.05M tokens
Input / 1M$10.00
Output / 1M$50.00

Runtime

Overall83.6
Verified (BenchLM)82
TTFT204.16 s
Throughput42 tok/s

Capability

Agentic86.4
Coding75
Knowledge85.4
Reasoning89.4
Multimodal & Grounded87.8
Instruction FollowingUnavailable
Math93.4

Claude Opus 5.5

Anthropic · rank 6

ProprietaryCodingReasoning
Context1M tokens
Input / 1M$4.00
Output / 1M$20.00

Runtime

Overall82.8
Verified (BenchLM)79
TTFT352.74 s
Throughput113 tok/s

Capability

Agentic66.3
Coding88
Knowledge86.1
Reasoning83.2
Multimodal & Grounded83.6
Instruction FollowingUnavailable
Math92

GPT-6.2 Sol

OpenAI · rank 7

ProprietaryReasoning
Context1.05M tokens
Input / 1M$2.00
Output / 1M$10.00

Runtime

Overall82.3
Verified (BenchLM)76
TTFTUnavailable
Throughput87 tok/s

Capability

AgenticUnavailable
CodingUnavailable
Knowledge84.9
Reasoning80.7
Multimodal & GroundedUnavailable
Instruction FollowingUnavailable
MathUnavailable

GPT-6.1 Sol

OpenAI · rank 8

ProprietaryCodingAgentsReasoning
Context1.05M tokens
Input / 1M$2.00
Output / 1M$10.00

Runtime

Overall81.1
Verified (BenchLM)73
TTFTUnavailable
Throughput43 tok/s

Capability

Agentic80.7
Coding72.8
Knowledge84.8
Reasoning86.6
Multimodal & Grounded87.8
Instruction FollowingUnavailable
Math93.8

Where does each model perform better?

Compare two models across seven capabilities. A gap means no score was published, not a score of zero.

Full comparison
Capability values for the two selected models
CapabilityClaude Fable 5.2Gemini 4 Neon
AgenticUnavailableUnavailable
Coding88Unavailable
Knowledge9483.1
Reasoning86.8Unavailable
Multimodal & Grounded89.3Unavailable
Instruction FollowingUnavailableUnavailable
Math93.892.5
Overall93.391.3

Evidence key: ObservedUnavailable

Cost simulator

What will my API usage cost?

Pick an API model and a daily workload. Token prices come from the catalog source.

Open the cost simulator
Estimated monthly API cost$45.06
Per conversation
$0.2048
Tokens / month
2.82M

8 messages per conversation, 1200 input and 400 output tokens per message, 22 active days. How we estimate workloads

Articles

Guides to model pricing, usage, and deployment.

All articles