anthropic
current · frontier · measured cost · verified 2026-08-21
Anthropic's most capable widely released model, GA 2026-06-09. Uses the Opus 4.7-generation tokenizer, which produces ~30% more tokens for the same text than pre-4.7 models — a real cost factor not visible in the per-token price.
Benchmark rows
Where the pass rate comes from Every row that names this model, with its harness and cost basis. Rows are harness + model combinations, not properties of the bare model.
Benchmark rows naming Claude Fable 5: benchmark, entry, pass rate, observed or usage-repriced attempt cost, cost basis and run date Benchmark Entry Pass $ / attempt Basis Run / read Artificial Analysis Coding Agent Index v1.4 Claude Code - Fable 5 (max) (with fallback) 67.0% $11.69 measured by source 2026-08-26
measured peers
Cost per solved task against its peers Measured rows only, ranked as on the calculator. Monthly figures at 200 tasks. Lower is better.
measured · the benchmark ran the model and observed this cost
MEASURED: cost per solved task, ranked 9 measured rows, ranked cheapest first. 1 DeepSeek V4 Pro $0.21; 2 Gemini 3.7 Flash $2.12; 3 GPT-5.6 Luna $2.65; 4 Muse Spark 1.2 $3.34; 5 Grok 4.5 $3.81; 6 GPT-5.6 Terra $4.50; 7 Qwen3.8 Max $5.30; 8 GLM 5.2 $15.49; 9 Claude Fable 5 $17.45. Monthly figures at 200 tasks. $ / SOLVED TASK · LOWER IS BETTER PASS $ / SOLVEDcost per solved task = cost per attempt ÷ pass rate. What it costs to get a task finished, not what a token costs. $ / MONTH$ / solved task × your monthly task volume. harness Codex · pass rate 43% · $0.09 per attempt 1 DeepSeek V4 Pro DeepSeek 43% $0.21 $42 compare › harness Opencode · pass rate 60% · $1.27 per attempt 2 Gemini 3.7 Flash Google 60% $2.12 $423 compare › harness Codex · pass rate 57% · $1.51 per attempt 3 GPT-5.6 Luna OpenAI 57% $2.65 $530 compare › harness Muse Code · pass rate 62% · $2.07 per attempt 4 ME Muse Spark 1.2 meta 62% $3.34 $668 compare › harness Grok Build · pass rate 64% · $2.44 per attempt 5 Grok 4.5 xAI 64% $3.81 $763 compare › harness Codex · pass rate 60% · $2.70 per attempt 6 GPT-5.6 Terra OpenAI 60% $4.50 $900 compare › harness Claude Code · pass rate 61% · $3.23 per attempt 7 QW Qwen3.8 Max qwen 61% $5.30 $1.1k compare › harness Claude Code · pass rate 43% · $6.66 per attempt 8 GLM 5.2 Z.ai 43% $15.49 $3.1k compare › harness Claude Code · pass rate 67% · $11.69 per attempt 9 Claude Fable 5 Anthropic 67% $17.45 $3.5k MEASURED: cost per solved task, ranked 9 measured rows, ranked cheapest first. 1 DeepSeek V4 Pro $0.21; 2 Gemini 3.7 Flash $2.12; 3 GPT-5.6 Luna $2.65; 4 Muse Spark 1.2 $3.34; 5 Grok 4.5 $3.81; 6 GPT-5.6 Terra $4.50; 7 Qwen3.8 Max $5.30; 8 GLM 5.2 $15.49; 9 Claude Fable 5 $17.45. Monthly figures at 200 tasks. harness Codex · pass rate 43% · $0.09 per attempt 1 DeepSeek V4 Pro $0.21 vs › $42/mo harness Opencode · pass rate 60% · $1.27 per attempt 2 Gemini 3.7 Flash $2.12 vs › $423/mo harness Codex · pass rate 57% · $1.51 per attempt 3 GPT-5.6 Luna $2.65 vs › $530/mo harness Muse Code · pass rate 62% · $2.07 per attempt 4 ME Muse Spark 1.2 $3.34 vs › $668/mo harness Grok Build · pass rate 64% · $2.44 per attempt 5 Grok 4.5 $3.81 vs › $763/mo harness Codex · pass rate 60% · $2.70 per attempt 6 GPT-5.6 Terra $4.50 vs › $900/mo harness Claude Code · pass rate 61% · $3.23 per attempt 7 QW Qwen3.8 Max $5.30 vs › $1.1k/mo harness Claude Code · pass rate 43% · $6.66 per attempt 8 GLM 5.2 $15.49 vs › $3.1k/mo harness Claude Code · pass rate 67% · $11.69 per attempt 9 Claude Fable 5 $17.45 $3.5k/mo Source: Artificial Analysis (artificialanalysis.ai) · verified 2026-08-21
Compare
Head to head Side-by-side pricing and cost per solved task with another current model.