Start with the reported Agentic Index, then check terminal and tool-use results against your workflow. Compare cost and completion time before choosing a provider.
The answer in 60 seconds
Claude Fable 5.1 (Max, Default Fallback) leads the reported Agentic Index at 57.9. Start with it for a capability-first shortlist, then compare alternatives against your workload, budget, and deployment requirements.
Using Artificial Analysis API benchmark data, cached for up to 24 hours. The source measurement date is not supplied. Selection rules reviewed .
Why this pick: Highest reported Agentic Index in this dataset; start here when benchmark capability is your priority.
Before choosing: Tool reliability and task cost depend on your prompts, tools, retries, and provider.
Also compare: Claude Opus 5 (Max).
Why this pick: Highest reported Agentic Index among the open-weight models in this dataset.
Before choosing: Open weights do not imply a permissive license or a fit for your hardware. Check the model license and memory requirements.
Also compare: GLM-5.3 (Max).
Why this pick: Highest reported τ²-Bench Telecom score among models with a result for that exact benchmark.
Before choosing: Coverage is incomplete. τ²-Bench Telecom results do not establish a winner for other tasks or benchmark versions.
Also compare: GLM 5V Turbo (Reasoning).
Source: Artificial Analysis model data. These are WhatLLM selection rules applied to published scores, not WhatLLM-run evaluations. Missing results are excluded from each comparison. Tied scores do not imply a performance difference. Read the methodology.
Only models with a reported Agentic Index enter this table. Equal scores are ordered by name. Individual task scores add context without changing the sort.
Cost per Intelligence Index task is a source benchmark metric. It is not the measured cost of successfully completing your workflow, and it does not affect this ranking.
Cached input pricing now matters for agent loops with repeated context, especially when prompts and tool traces get large.
Agentic fit lab
Task
Agentic Index preference
Models below this target rank lower. It does not certify autonomy or reliability.
Minimum context
Optimize for
Shared links preserve these settings and exact model variants. Recommendations are recalculated from current data and may change.
Fit is a WhatLLM screening score, not a benchmark or a probability of success. Every result has an observed Agentic Index, meets the minimum advertised context window, and respects the open weights filter when selected. Context capacity does not measure retrieval accuracy.
For codebase agent, the base weights are Agentic Index 34%, Intelligence Index 12%, Coding Index 22%, terminal benchmarks 16%, τ-Bench 4%, GDPval 4%, context 4%, cost 2%, and response time 2%.
Missing metrics earn no ranking credit. Index and percentage scores are scaled to 0–1; GDPval Elo uses a WhatLLM 900–1800 scale. Cost and time are normalized across the current dataset. Your optimization preference adds weight to the selected measure, and models below your Agentic Index target receive a penalty.
Cost per task means an Artificial Analysis Intelligence Index evaluation task. It is not cost per successful agent task. Check the exact model variant, provider, and results on your own workload before choosing. Sources: Artificial Analysis and WhatLLM methodology.
Fit
66
Agentic
50.2
Cost / task
$1.99
Response
2.2m
Context
1.0M
Start your evaluation here: GPT-5.6 Sol (Max) leads this workload’s weighted ranking, with an observed Agentic Index of 50.2 and 1.0M context. It is below your preferred Agentic Index target.
AA evaluation task cost exceeds $1.50; check costs on your workload.
Check this model’s pricing and providersValue route
OpenAI
Fit
39
Task cost
$0.085
Response
56.5s
Benchmark cost and response time describe AA evaluations; validate your own agent loop.
Fast route
OpenAI
Fit
59
Task cost
$0.808
Response
39.1s
Benchmark cost and response time describe AA evaluations; validate your own agent loop.
Open route
Alibaba
Fit
58
Task cost
$0.372
Response
50.8s
Benchmark cost and response time describe AA evaluations; validate your own agent loop.
Frontier map
12 meet the Agentic Index target
135 with cost/task
GPT-5.6 Sol (Max)
OpenAI
Agentic
50.2
Task cost
$1.99
Agentic Index
57.9
Anthropic
Agentic Index
56.5
Anthropic
Agentic Index
56
Alibaba
Sorted by reported Agentic Index. Cost and time are separate source metrics, not ranking weights or guarantees for your workflow. Compare only the same benchmark version and unit.
| Rank | Model | Agentic | Cost / task | Terminal-Bench | τ-Bench | GDPval-AA | Response |
|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.1 (Max, Default Fallback) Anthropic | 57.9 | $7.63 | N/A | N/A | 1756GDPval-AA Elo | 4.4m |
| 2 | Claude Opus 5 (Max) Anthropic | 56.5 | $5.86 | N/A | N/A | 1724GDPval-AA Elo | 1.3m |
| 3 | Qwen3.8 Max (0902) Alibaba | 56 | $5.41 | N/A | N/A | 1671GDPval-AA Elo | 1.2m |
| 4 | Muse Spark 1.3 (Max) Meta | 55.5 | $1.60 | N/A | N/A | 1681GDPval-AA Elo | 1.1m |
| 5 | Qwen3.8-Flash-Next Alibaba | 53.6 | $0.372 | N/A | N/A | 1632GDPval-AA Elo | 50.8s |
| 6 | GLM-5.3 (Max) Z AI | 53.1 | $2.01 | N/A | N/A | 1651GDPval-AA Elo | 41.2s |
| 7 | Grok 4.6 (High) SpaceXAI | 53 | $1.48 | N/A | N/A | 1622GDPval-AA Elo | 55.6s |
| 8 | GPT-6 Astra (Max) OpenAI | 51 | $3.26 | N/A | N/A | 1542GDPval-AA Elo | 6.5m |
| 9 | GLM-5.3-Flash Z AI | 50.9 | $0.253 | N/A | N/A | 1644GDPval-AA Elo | 43.8s |
| 10 | Claude Fable 5 (Max, Opus 4.8 Fallback) Anthropic | 50.7 | $8.75 | 63%Terminal-Bench Hard | 99%τ²-Bench Telecom | 1613GDPval-AA Elo | 2.1m |
If your agents browse, call APIs, run tools, or plan several steps ahead, start here rather than on a generic leaderboard. The best agentic model is the one that stays reliable under execution at a cost and response-time profile you can actually ship.
Once you have a shortlist, open the finalists on Compare and validate whether the provider you want can deliver the right price and latency profile.
Agentic performance overlaps with coding and long-context performance, but it is not the same thing. Use the related pages below if your use case is specialized around software work, model routing, or large-document reasoning.
Ranking library
Focused rankings for the decisions engineers actually make.
Claude Fable 5.1 (Max, Default Fallback) leads the reported Agentic Index at 57.9. Start with it for a capability-first shortlist, then compare alternatives against your workload, budget, and deployment requirements.
Agentic models can maintain plans across many steps, call tools reliably, follow complex instructions, and recover gracefully when a workflow hits an unexpected state.
Not always. Coding strength helps with tool-writing and structured output, but agentic performance also requires strong planning, tool orchestration, and error recovery.
Use benchmarks to get a shortlist, then run the finalists on the actual tasks your agent will perform. Production testing on real workflows is the only way to make the final call.
Sequential model and tool calls add to completion time. Measure the full workflow, including tool latency, retries, and failures; token generation speed alone does not predict completion time.
The table sorts by reported Agentic Index, excluding models without it. Terminal-Bench, τ-Bench, and GDPval-AA provide task-specific evidence. Compare the same benchmark version and validate finalists on your own tasks.