What's new
1h data Recomputed benchmark-weighted quality scores 1h data Synced Chatbot Arena benchmark track 1h data Updated speed measurements 1h data Validated official pricing snapshots 1h data Pulled latest OpenRouter price index 10h jobs Jobs market snapshot refreshed 25 May digest OpenGuardrails: An Open-Source Context-Aware AI Guardrails Platform 25 May digest Published the 2026-05-25 daily digest 25 May digest Workbench Launches Open Source BullMQ Dashboard For Node Backends 24 May digest SpecBench Tests Reward Hacking In Long Horizon Coding Agents 1h data Recomputed benchmark-weighted quality scores 1h data Synced Chatbot Arena benchmark track 1h data Updated speed measurements 1h data Validated official pricing snapshots 1h data Pulled latest OpenRouter price index 10h jobs Jobs market snapshot refreshed 25 May digest OpenGuardrails: An Open-Source Context-Aware AI Guardrails Platform 25 May digest Published the 2026-05-25 daily digest 25 May digest Workbench Launches Open Source BullMQ Dashboard For Node Backends 24 May digest SpecBench Tests Reward Hacking In Long Horizon Coding Agents

Anthropic

Claude Opus 4.8

AA puts Opus 4.8 at 61 on Intelligence Index, #2 on AA-Omniscience with 46.6% accuracy and 35.9% hallucination rate; Kilo has it #3 coding at 67.6%.

80 Floor score

confidence: high

80 Floor Capability

Finished benchmark shape.

60 Operating Envelope

Strong DSWE pass rate, but high average cost and token burn pull the operating score down.

88 Frontier Reality

AA Index 61 / AA-Omniscience #2

82 Research Trust

Best clean Anthropic row in the current set, with hallucination and accuracy evidence available.

Current signal

AA Index 61 / AA-Omniscience #2

Evidence completeness

82%

6 score dimensions represented.

Calculation

Score components

80 Accuracy + grounding

Correct, grounded answers on ordinary verifiable tasks.

80 Instruction discipline

Follows exact user constraints and output formats.

86 Honesty + reliability

Avoids unsupported confidence and remains stable across runs.

84 Reasoning floor

Handles non-flashy reasoning without brittle failures.

78 Agent/coding

Completes tool, coding, and agent-style tasks when evidence exists.

62 Operating envelope

Cost, speed, token burn, verbosity, and friction.

Caveat

What to watch

Strongest clean Anthropic row for floor-style comparison. Cost is high, so operating envelope drags the meta score down.

Meaning

How to read the score

Each score is a 0-100 saturation scale. A score of 100 means that benchmark lane is complete for this model. Floor Capability is the finished benchmark shape; Operating Envelope, Frontier Reality, and Research Trust are provisional tabs that need prompt-pack backfill. Confidence is evidence quality, not another model grade. Evidence % is source coverage, not capability.

Citations

Benchmark sources

Artificial Analysis Intelligence IndexAA-OmniscienceHumanity's Last ExamIFBenchTerminal-Bench HardKilo Bench