Data as of November 13, 2026 · How the score is built
Flint
Decision readingFlint is tracked, but not publicly ranked yet. The profile exposes 5 sourced benchmark rows and leaves unsupported fields blank until a published record exists.
Email me when its price, rank or availability changes.
Capability
Unranked
field median 49.4Not eligible for a public rank
Price
Self-hosted; infrastructure cost varies
input median $0.60No comparable first-party hosted token rate
Speed
Not measured
field median 92 tok/sTime to first token not measured
Context
256Ktokens
field median 256,000Maximum output length is tracked separately
Decision snapshot
How much published evidence this model has, where that evidence is strongest, and what to validate before choosing it. It has no public rank, so no field position is shown.
Strongest published evidence
Instruction Following ranks #3. A well-rounded choice across a range of tasks.
Validate before choosing
5 published rows leave some tracked benchmark slots empty. No comparable first-party API token rate is published.
Source-linked · 5 displayable benchmark rows
Is Flint right for your job?
Its place on the LLM Selector’s job rankings, the same order the job pages publish. Describe your own job to see where it stands for that.
Where Flint ranks by benchmark score:
We keep what you type here, drafts included, for 180 days, and after that only the job and requirements read from it. Nothing is tied to your IP address or account. Leave out names and other personal details. Privacy
Category score record
Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.
| Category | Score | Rank | Percentile | Weight | Benchmarks | Evidence |
|---|---|---|---|---|---|---|
| AgenticWeight 22%1 benchmarkVerified | Score pending | 1 benchmark | Verified | |||
| CodingRank Not rankedWeight 20%2 benchmarksVerified | 62.0 | 2 benchmarks | Verified | |||
| ReasoningWeight 17%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| MultimodalWeight 12%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| KnowledgeWeight 12%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| MultilingualWeight 7%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| Inst. FollowingRank #3 of 157Percentile 99thWeight 5%2 benchmarksVerified | 94.7 | 2 benchmarks | Verified | |||
| MathWeight 5%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured |
5 of 919 tracked benchmark slots have displayable evidence · bars run 0–100
Coverage detailsHow much of this is verified
Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.
- Agentic1/1 verified
- Coding2/2 verified
- ReasoningNot measured
- MultimodalNot measured
- KnowledgeNot measured
- MultilingualNot measured
- Inst. Following2/2 verified
- MathNot measured
Benchmark ledger
Every result says who ran the test. For this model, Recursive ran all 6. Coding opens by default. The tick marks the best source-verified result in the catalog; provisional leaders do not set the reference.
Coding2 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| SWE-bench VerifiedSoftware Engineering Benchmark Verified | 74.2% | Best verified: Claude Opus 5 · 96% | Gap21.8 behind | Reported by Recursive | Shown only |
| LiveCodeBench v6 | 80.3% | Best verified: Sakana Fugu-Ultra · 93.2% | Gap12.9 behind | Reported by Recursive | Shown only |
Agentic1 row
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| Terminal-Bench 2.1 | 54.9% | Best verified: SWE-2 · 92.8% | Gap37.9 behind | Reported by Recursive | Weighted benchmark9% ref. weight |
Inst. Following2 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| IFBenchInstruction Following Benchmark | 89.2% | Best verified: Flint · 89.2% | GapHighest listed | Reported by Recursive | Weighted benchmarkWeighted 70% |
| IFEvalInstruction-Following Eval | 95.1% | Best verified: LFM2.5-24B-A2B · 96.8% | Gap1.7 behind | Reported by Recursive | Shown only |
Other lab-reported results1 rowAnother metric, split or setting, or a test without a page
| Benchmark | Score | On a 0–100 scale | Who ran it | Use in ranking |
|---|---|---|---|---|
| SWE-bench Pro v2Resolved rate · Edition v2, which differs from the catalog’s SWE-bench Pro | 89.7% | Reported by Recursive | Shown only |
Outlined bars are lab-reported. Filled bars are benchmark-run or independent. Bars run 0–100; the dark tick marks the best source-verified value.
Lineage
The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
Base entry
Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- Weights model ID
- RecursiveSI/flintRecursive Flint model card
- Context window
- 256KRecursive Flint model card
- Maximum output
- Not sourced yet
- Knowledge cutoff
- Not sourced yet
- Input modalities
- textRecursive Flint model card
- Output modalities
- textRecursive Flint model card
- Parameters
- 39.8B total, approximately 2.5B active per tokenHugging Face safetensors metadata: 39,785,287,456 BF16 parameters
- Availability
- Hugging Face · self-hosted inferenceRecursive Flint model card
- Cloud regions
- Not tracked yet
- Lifecycle
- activeRecursive Flint model card
- API capabilities
- Tool calling, structured outputs, and batch support are not tracked yet
- Prompt caching
- Not documented in the pricing recordRecursive Flint model card
- Self-host
- Flint needs ~24GB VRAM (1× NVIDIA RTX 4090 (24GB)).
- Rate limits
- Not tracked yet
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
We track Flint, but the public leaderboard excludes this profile until enough non-generated benchmark coverage is available. Only published rows appear above.
Flint is a open weight model with a 256K context window. It uses an explicit reasoning mode, which can improve complex problem solving while adding latency and token use.
Open Apache 2.0 BF16 weights on Hugging Face (about 80 GB) with an SGLang serving kit for one B300 or H200 per replica; Recursive also ships the rec engine for Apple Silicon and lists llama.cpp, Ollama, LM Studio, vLLM and Unsloth support. The card documents a 262,144-token context; the launch post says up to 512K. Recursive's hosted API terms are separate and no token price is published.
Recursive Superintelligence released Flint on October 21, 2026: an Apache 2.0, text-only mixture-of-experts model for coding agents with 39.8B total and about 2.5B active parameters, trained from scratch. The model card reports Terminal-Bench 2.1 (Terminus 2, web-retrieved solutions counted as failures), SWE-bench Verified (mini-swe-agent, no internet), LiveCodeBench v6, IFBench and IFEval at temperature 1.0, top-p 0.95, top-k 20, averaged over at least five attempts for the agentic runs. Those five results are stored for the BF16 checkpoint. The card also reports a SWE-bench Pro v2 result, an edition that differs from the catalog's SWE-bench Pro lane, so it stays unmapped and is listed in the benchmark ledger. The NVFP4 build is not a catalog row. The row is display-only and outside rankings.
5 of 919 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Its strongest eligible category is Instruction Following at #3. a well-rounded choice across a range of tasks.
Last updated November 13, 2026. Runtime fields remain blank until a sourced snapshot exists.
Deployment options
Self-host and provider-specific paths stay separate from benchmark evidence so operating constraints are visible before a score becomes the whole decision.
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Questions
How does Flint perform overall in AI benchmarks?
Flint has 5 source-displayable benchmark rows, but it does not qualify for a public overall rank. The available rows remain visible by category without being converted into a site-wide score. Missing evidence stays blank instead of being estimated from an earlier model.
Is Flint good for coding and programming?
Flint has source-displayable benchmark coverage for coding and programming, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is Flint good for agentic tool use and computer tasks?
Flint has source-displayable benchmark coverage for agentic tool use and computer tasks, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is Flint good for instruction following?
Flint ranks #3 out of 157 eligible models for instruction following, with a public category score of 94.7/100. That places it in the current top ten for this category. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Flint open source?
Flint is an open-weight model from Recursive. Its weights can be downloaded for local or hosted deployment, subject to the published license. Open weight does not automatically mean open source: training data and training code may remain private, and commercial restrictions can still apply.
Does Flint have full benchmark coverage on BenchLM?
No. Flint currently has 5 source-displayable rows. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of Flint?
Flint has a documented context window of 256K. That figure is the maximum combined prompt and retained-conversation space reported for this exact model; it is not the maximum output length. The profile keeps output limits separate because providers often publish those limits independently.