Designing
Indian Workflow Index
69
Support, finance, legal, sales, document, and multilingual tasks scored by outcome, not trivia recall.
Task completion
Hindi-English robustness
Citation quality
Escalation handling
Leaderboards
Rankings for workflows, agents, cost, and latency.
These are prototype tracks for a public Indian AI benchmarking layer. The goal is not a single universal score; it is a set of buyer-relevant views with visible methodology.
Prototype
Agentic Reliability Index
53
Measures tool-use planning, recovery, browser discipline, terminal usage, and whether agents can verify their own work.
Planning
Tool calls
Recovery
Verification
Live draft
Cost Efficiency Index
75
Converts provider pricing into cost per completed workflow using cache hit assumptions, output length, and retries.
Input cost
Output cost
Retry rate
Batch/cache leverage
Collecting
Latency and Throughput Index
74
Separates time-to-first-token, output tokens per second, queueing behavior, and provider variance.
TTFT
Tokens/sec
P95 latency
Availability
Methodology before rankings.
Every public number should trace back to task provenance, model settings, sample count, scoring rubric, and the failure cases the score hides.
Read methodologyPrototype model comparison
Frontier reasoning modelBest quality and recovery
Fast mid-tier modelBest default for support
Open-weight local modelUseful when privacy/control beats maximum benchmark quality.
Small routing modelCheap classifier/router
Agentic Reliability Formula
A trace-derived index for coding, browser, support, and security agents.
Trace-derived Agentic Reliability Index for comparing coding, browser, and support agents by completion, state proof, recovery, tool/policy correctness, and cost-latency discipline. The formula keeps completion, state proof, recovery, tool discipline, and operating cost visible instead of hiding them behind a single rank.
4
Suites
16
Traces
5
Weights
30%
Task completion
Measures whether the agent produced the accepted workflow result.
25%
Evidence and state verification
Rewards proof that the browser, codebase, support policy, or tool state actually reached the target.
20%
Recovery behavior
Separates agents that recover from validation errors, failed tools, and partial state from agents that simply stop.
15%
Tool and policy correctness
Captures tool discipline, policy adherence, escalation quality, and repository hygiene.
10%
Cost and latency discipline
Prevents slow or expensive agents from ranking well unless quality justifies the operating cost.
Model classIndexCompleteProofRecoverCost/latency
Frontier reasoning model7280667655
Fast mid-tier model6569596283
Open-weight local model4749444467
Small routing model3732323094
Prototype run matrix
How a buyer should read benchmark tradeoffs
Model classWorkflowAgenticCostLatencyInterpretation
Frontier reasoning model88795060Best quality and recovery, expensive for high-volume routing.
Fast mid-tier model76688184Best default for support, extraction, and most back-office work.
Open-weight local model61527458Useful when privacy/control beats maximum benchmark quality.
Small routing model52409492Cheap classifier/router, not a final-answer model for hard cases.