Recursive Flint
Last verified October 23, 2026 · Recursive pricing ↗<$0.0001/1M input · <$0.0001/1M output · 262K context · Recursive
Count Tokens
—
Tokens
—
Input Cost
—
Output Cost
Estimate Monthly Cost
Monthly Cost Estimator
Quick:
<$0.0001/mo
Pick a preset above or enter custom usage
Alternatives to Recursive Flint
Pricing Details
RecursiveOpen-weight (Apache 2.0) coding model from Recursive Superintelligence, released 2026-10-21. MoE: 39.8B total, ~2.5B active (49 layers, 256 routed experts, 6 active + 1 shared). Context 262,144 (launch blog says up to 512K, recommends under 256K). NO PAID PER-TOKEN API as of 2026-10-23: OpenRouter recursive/flint-v1.0:free is $0/$0 served by Recursive in BF16; the paid ID recursive/flint-v1.0 has zero endpoints; no Together/Fireworks/DeepInfra listing found. OpenRouter free limits: 20 RPM; 50 requests/day under $10 lifetime credit purchases, 1,000/day at $10+. Recursive mentions a hosted API with separate terms, no published price. Recommended output budget 32,768+ tokens on hard tasks; reasoning mandatory on OpenRouter. Builds: BF16 79.59 GB, MLX INT4 22.58 GB, NVFP4 GGUF 22.86 GB (release quant, within 0.7 pts of 16-bit on average), INT4 Q4_1 GGUF 25.99 GB. LM Studio lists 24 GB RAM minimum. Recursive-claimed speed 125+ tok/s on M4 Pro MacBook Pro, 150+ on 64GB RTX Spark, at short contexts. Recursive-reported (HF card chart, BF16): Terminal-Bench 2.1 54.9, 'SWE-bench Pro v2' 89.7 (Recursive's in-house measurement, NOT public SWE-bench Pro), SWE-bench Verified 74.2, LiveCodeBench v6 80.3, IFBench 89.2, IFEval 95.1. Dense Qwen 3.8 beats it on TB 2.1 (73.0) and Pro v2 (94.9). No Artificial Analysis page as of 2026-10-23.
Input / 1M tokens
$0
Output / 1M tokens
$0
Context Window
262K
Max Output
33K
Price History
Launched at current price on 2026-10-21. No price changes recorded since.
Frequently Asked Questions
Common questions about Recursive Flint pricing and usage