RecursiveReleased Oct 21, 2026
NewOpen weights
Flint
Flint doesn't have a Composite IQ yet. A score needs measured results in at least two of the seven dimensions, and none have published results so far.
Composite IQ ?
—
Effective cost
Not priced
No standard API rate
Speed
Not measured
No speed data yet
Context
Unverified
Not in publisher docs
Evidence
0of 7 dims
0 of 39 benchmarks
IQ profile
Seven dimensions, averaged equally into Composite IQ
Profile not available yetThe radar appears once at least two dimensions have measured benchmark results.
MeasuredEstimated
#1 in dimensionTop-10 median
IQ dimensions
Ranks are among models scored in each dimension. Ranges are model-based estimates, not calibrated confidence intervals.
DimensionIQPosition · 70–160RankEvidencevs Top 10
FlintTop-10 median#1 in dimension100 = average human
Benchmark results
0 of 39 Composite IQ benchmarks have a published result for Flint. Benchmarks without a result stay listed so gaps are visible.
Abstract Reasoning IQ 99 · estimatedNo published results0/5
No published results yet for Abstract Reasoning. The 99 shown is estimated from Flint's other dimensions and will be replaced as results come in. How estimates work
ARC-AGI-3ARC-AGI-2ARC-AGI-1Chess PuzzlesMystery Game Puzzles
Mathematical Reasoning no scoreNo published results0/5
No published results yet for Mathematical Reasoning. How estimates work
FrontierMath Tier 4FrontierMath Tier 1-3ProofBench 1.1MathArenaAIME
Academic Reasoning IQ 100 · estimatedNo published results0/6
No published results yet for Academic Reasoning. The 100 shown is estimated from Flint's other dimensions and will be replaced as results come in. How estimates work
Humanity's Last ExamGPQA DiamondCritPtSciCodeMMLU-ProMMMU-Pro
Programmatic Reasoning no scoreNo published results0/6
No published results yet for Programmatic Reasoning. How estimates work
LiveCodeBenchIOI V2Terminal-Bench 4.0ProgramBenchSWE-rebenchMirrorCode
Computer Use no scoreNo published results0/5
No published results yet for Computer Use. How estimates work
BrowseCompOSWorld-VerifiedMCP AtlasArena.ai Agent ArenaAgents' Last Exam
Reliability no scoreNo published results0/6
No published results yet for Reliability. How estimates work
SimpleQA VerifiedaaOmniscienceBullshitBench v2IFBenchAA Long Context Reasoning v1.1FACTS Grounding
Knowledge Work no scoreNo published results0/6
No published results yet for Knowledge Work. How estimates work
AA-Briefcase v1.1GDPval-AA v2.1AutomationBench-AAGDP.pdfAA-AnalystAgentEnterpriseOps-Gym-AA
Release & availability
Facts from the publisher's announcement and documentation.
- Released
- Developer
- Recursive
- Access
- Apache 2.0 open weights in the publisher RecursiveSI/flint repository; local serving support and hosted OpenRouter access.
- Weights
- Open weightsDownloadable and self-hostable
- Inputs
- text
- Context
- Not publishedThe publisher recommends staying within 256K tokens; its release page states up to 512K, while the BF16 card documents a 262,144-token shared prompt/output configuration.
- API price
- No per-token rateStandard hosted token pricing is unverified; the OpenRouter free offering is not a standard paid rate.
What Recursive says
- Mixture-of-experts model for coding and tool use, with 39.8B total and approximately 2.5B active parameters.
Limitations
- Context capacity depends on checkpoint and runtime.
- Publisher agent-harness and local throughput claims are not imported as independent benchmark measurements.
Publisher statements, not AI IQ measurements.
About this score
How much of Flint's IQ is measured, and how the rest is estimated.
0/ 39benchmark results
0/ 7dimensions measured
2estimated
0/5Abstract
0/5Math
0/6Academic
0/6Coding
0/5Computer Use
0/6Reliability
0/6Knowledge Work
Bar height = share of benchmarks with resultsEstimatedNo score
Measured results are shown as published
Every score in the benchmark tables comes from its source unchanged. Composite IQ averages all seven dimensions equally.
Estimates fill the gaps
Abstract Reasoning and Academic Reasoning have no published results yet, so their IQ is estimated from Flint's other dimensions.
A missing result is not a zero
Benchmark versions and evaluation harnesses can differ between sources, so small gaps between models may not be meaningful.
Compare Flint
Side-by-side pages for the models it is most often weighed against.
#1 overall
gpt-6.1-sol
IQ141Effective cost$6.25
US (Other) #1
Beam
IQ114
#2 overall
opus-5.5
IQ141Effective cost$35.41
#3 overall
Gemini 4 Argon
IQ140
#4 overall
gpt-6-astra
IQ140Effective cost$33.91
#5 overall
fable-5.1
IQ137Effective cost$90.23
FlintComparison model
Frequently asked questions
Quick answers about Flint, drawn from the same data as the rest of this page.
How AI IQ scores models →Get this data from the API →- What is the IQ of Flint?
- Flint does not yet have an estimated IQ. AI IQ publishes a composite score once a model has source-backed benchmark results in at least two of its seven capability dimensions.
- What is Flint best at?
- There isn't any published benchmark data to judge that yet. None of the seven dimensions AI IQ scores has a measured result for Flint, so its strengths and weaknesses can't be compared with other models.
- What is the EQ of Flint?
- Flint doesn't have an EQ score yet. EQ (emotional reasoning) is calculated from EQ-Bench 4 and AttuneBench, and neither has published a result for Flint. AI IQ doesn't estimate EQ from other benchmarks, so the score appears once one of those results is published.
- Which benchmarks has Flint been tested on?
- None yet. Flint has no published results on the 39 benchmarks behind AI IQ's Composite IQ. AI IQ only uses published, source-backed results, so its profile fills in as evaluations of this model appear.
- How much does Flint cost?
- AI IQ doesn't have a verified standard per-token API price for Flint, so it can't calculate an effective cost. This is common for models offered only through subscriptions, restricted-access programs, or self-hosting.
- Is Flint open source?
- Yes — Flint is an open-weights model: its trained weights are published, so it can be downloaded and self-hosted.
- Who developed Flint?
- Flint was developed by Recursive, based in the United States.
- When was Flint released?
- Flint was publicly released on October 21, 2026.
- How large is the context window of Flint?
- AI IQ hasn't verified an input context limit for Flint from its developer's documentation, so none is listed rather than guessing.
- How fast is Flint?
- AI IQ doesn't have independent speed measurements for Flint yet. Speed comes from independent tests of a model's hosted API, which aren't available for every model — especially new, restricted-access, or self-hosted ones.