Raw token price is only the starting point. AICodingPricing connects coding benchmark evidence, API pricing, cache rules, context limits, model caveats, and task-cost assumptions so you can choose a model for the work you are actually running.
A model can lead one benchmark, cost less per token, or offer a larger context window, but still be the wrong choice if it retries more often, produces longer trajectories, lacks exact source-backed pricing, or has weak evidence for your task type.
Use the filters to compare models by workflow, not by hype. A filter is not an absolute ranking. It is a lens over source-backed fields, confidence labels, and caveats.
Candidates for multi-step coding-agent loops need public coding evidence, visible retry caveats, and source-backed token pricing where available.
This is a shortlist label, not a universal winner. Estimate task cost before committing to a default model.Frontend generation needs coding evidence plus human inspection of UI output, follow-up rate, and implementation cost.
Preference or coding scores do not guarantee pixel-perfect UI. Treat the label as a workflow lens.Repo-level refactor and code review need editing evidence, context caution, and visible source freshness before cost comparison.
Advertised context is not the same as reliable long-context repo editing.Low-cost automation starts with token price, then checks benchmark coverage, retry rate, cache behavior, and cleanup cost.
Do not call a model cheapest overall unless task assumptions are visible.Kimi, Qwen, DeepSeek-style rows can be useful when the page separates pricing, benchmark coverage, context, and unknown fields.
Partial evidence is acceptable. Fake certainty is not.Long-context candidates must show source-backed context values or mark missing context as not disclosed.
Large context windows do not prove reliable repo-level editing.Each row shows the model, provider, benchmark evidence, input price, output price, cache price, context window, speed signal when available, best-for labels, caveat, source, last checked date, confidence, and data status. If a field is unknown, the reason stays visible.
| Model | Benchmark evidence | Input / Output | Cache / Context | Best for | Caveat / source | Task cost |
|---|---|---|---|---|---|---|
| Claude Opus 4.5 Anthropic partial | 76.80% SWE-bench Verified · medium Benchmark source: SWE-bench Leaderboards · checked 2026-06-02 Captured source names the row as Claude 4.5 Opus — high reasoning; verify exact API model alias before display. | Input: $5 Output: $25 | $0.5 read; $6.25 5m write not disclosed | Coding agentsRepo refactorCode review | Strong SWE-bench evidence in captured source, but expensive output pricing; do not present as universal best model. Pricing source: Anthropic Claude API pricing · checked 2026-06-02 · confidence high | Calculate This Model’s Task Cost |
| Claude Sonnet 4.5 Anthropic partial | 71.40% SWE-bench Verified · medium Benchmark source: SWE-bench Leaderboards · checked 2026-06-02 Captured source names the row as Claude 4.5 Sonnet — high reasoning; verify exact API model alias before display. | Input: $3 Output: $15 | $0.3 read; $3.75 5m write not disclosed | Coding agentsFrontend generationRepo refactorTest generation | Good candidate for default coding workflow shortlist, but scenario label must cite benchmark and price fields rather than claim overall superiority. Pricing source: Anthropic Claude API pricing · checked 2026-06-02 · confidence high | Calculate This Model’s Task Cost |
| Claude Haiku 4.5 Anthropic partial | 66.60% SWE-bench Verified · medium Benchmark source: SWE-bench Leaderboards · checked 2026-06-02 Captured source names the row as Claude 4.5 Haiku — high reasoning; verify exact API model alias before display. | Input: $1 Output: $5 | $0.1 read; $1.25 5m write not disclosed | Low-cost automationTest generationBug fixing | Lower token price does not guarantee lower task cost if retry rate rises; calculator must expose retry assumptions. Pricing source: Anthropic Claude API pricing · checked 2026-06-02 · confidence high | Calculate This Model’s Task Cost |
| GPT-5.4 mini OpenAI partial evidence | not publicly benchmarked SWE-bench Verified · low Benchmark source: SWE-bench Leaderboards exact alias recheck · checked 2026-06-05 Exact SWE-bench row for GPT-5.4 mini was not verified. A public SWE-bench row exists for a different OpenAI Mini alias, so this row intentionally does not attach its numeric value. | Input: $0.75 Output: $4.5 | $0.075 read; write not disclosed not disclosed | Low-cost automationTest generation | OpenAI API pricing is source-backed for GPT-5.4 mini standard short-context pricing. Exact public coding benchmark for this alias is not verified, so benchmark evidence is marked not publicly benchmarked. Pricing source: OpenAI API pricing · checked 2026-06-02 · confidence high | Calculate This Model’s Task Cost |
| Gemini 3 Flash — high reasoning partial | 75.80% SWE-bench Verified · medium Benchmark source: SWE-bench Leaderboards · checked 2026-06-02 Captured benchmark row is Gemini 3 Flash — high reasoning; exact Google API pricing row for this exact model alias was not captured. | Input: not disclosed Output: not disclosed | not disclosed not disclosed | Coding agentsLong contextFrontend generation | Strong captured SWE-bench result, but price/context must remain unknown until exact Gemini model docs are mapped. Pricing source: Gemini Developer API pricing · checked 2026-05-28 · confidence low | Calculate This Model’s Task Cost |
| DeepSeek V4 Flash DeepSeek partial | not_publicly_benchmarked SWE-bench / Aider · low Benchmark source: DeepSeek API Docs + public benchmark sources checked · checked 2026-05-28 Captured benchmark sources mention DeepSeek V3.2 / R1 variants, not exact DeepSeek V4 Flash. | Input: $0.14 Output: $0.28 | $0.0028 read; write not disclosed 1,000,000 tokens | Low-cost automationChinese coding workflowLong context | Excellent token price and context signal, but exact public coding benchmark row for V4 Flash was not captured; mark coding evidence as incomplete. Pricing source: DeepSeek API Docs — Models & Pricing · checked 2026-05-28 · confidence high | Calculate This Model’s Task Cost |
| DeepSeek V4 Pro DeepSeek partial | not_publicly_benchmarked SWE-bench / Aider · low Benchmark source: DeepSeek API Docs + public benchmark sources checked · checked 2026-05-28 Captured benchmark sources mention DeepSeek V3.2 / R1 variants, not exact DeepSeek V4 Pro. | Input: $0.435 Output: $0.87 | $0.0036 read; write not disclosed 1,000,000 tokens | Chinese coding workflowLong contextRepo refactor | Pricing/context are source-backed; coding benchmark evidence for the exact V4 Pro model still needs source verification. Pricing source: DeepSeek API Docs — Models & Pricing · checked 2026-05-28 · confidence high | Calculate This Model’s Task Cost |
| Kimi K2.5 — high reasoning Moonshot AI / Kimi partial | 70.80% SWE-bench Verified · medium Benchmark source: SWE-bench Leaderboards · checked 2026-06-02 Captured source names Kimi K2.5 — high reasoning. | Input: not disclosed Output: not disclosed | not disclosed not disclosed | Chinese coding workflowCoding agentsRepo refactor | Useful Chinese coding workflow candidate, but price/context must remain unknown until exact Moonshot pricing/model docs are captured. Pricing source: Kimi API Platform pricing index · checked 2026-05-28 · confidence low | Calculate This Model’s Task Cost |
| Qwen3 235B A22B Alibaba Cloud / Qwen partial | 59.6% Aider polyglot coding benchmark · medium Benchmark source: Aider LLM Leaderboards · checked 2026-05-28 Captured source row: Qwen3 235B A22B diff, no think, Alibaba API. | Input: not disclosed Output: not disclosed | not disclosed not disclosed | Chinese coding workflowLow-cost automation | Benchmark-backed partial row only. Do not show price until exact Alibaba Cloud model pricing is captured. Pricing source: Alibaba Cloud Model Studio pricing search result · checked 2026-05-28 · confidence low | Calculate This Model’s Task Cost |
Strong SWE-bench evidence in captured source, but expensive output pricing; do not present as universal best model.
Pricing source: Anthropic Claude API pricing · checked 2026-06-02 · confidence high
Benchmark source: SWE-bench Leaderboards · checked 2026-06-02 · confidence medium
Good candidate for default coding workflow shortlist, but scenario label must cite benchmark and price fields rather than claim overall superiority.
Pricing source: Anthropic Claude API pricing · checked 2026-06-02 · confidence high
Benchmark source: SWE-bench Leaderboards · checked 2026-06-02 · confidence medium
Lower token price does not guarantee lower task cost if retry rate rises; calculator must expose retry assumptions.
Pricing source: Anthropic Claude API pricing · checked 2026-06-02 · confidence high
Benchmark source: SWE-bench Leaderboards · checked 2026-06-02 · confidence medium
OpenAI API pricing is source-backed for GPT-5.4 mini standard short-context pricing. Exact public coding benchmark for this alias is not verified, so benchmark evidence is marked not publicly benchmarked.
Pricing source: OpenAI API pricing · checked 2026-06-02 · confidence high
Benchmark source: SWE-bench Leaderboards exact alias recheck · checked 2026-06-05 · confidence low
Strong captured SWE-bench result, but price/context must remain unknown until exact Gemini model docs are mapped.
Pricing source: Gemini Developer API pricing · checked 2026-05-28 · confidence low
Benchmark source: SWE-bench Leaderboards · checked 2026-06-02 · confidence medium
Excellent token price and context signal, but exact public coding benchmark row for V4 Flash was not captured; mark coding evidence as incomplete.
Pricing source: DeepSeek API Docs — Models & Pricing · checked 2026-05-28 · confidence high
Benchmark source: DeepSeek API Docs + public benchmark sources checked · checked 2026-05-28 · confidence low
Pricing/context are source-backed; coding benchmark evidence for the exact V4 Pro model still needs source verification.
Pricing source: DeepSeek API Docs — Models & Pricing · checked 2026-05-28 · confidence high
Benchmark source: DeepSeek API Docs + public benchmark sources checked · checked 2026-05-28 · confidence low
Useful Chinese coding workflow candidate, but price/context must remain unknown until exact Moonshot pricing/model docs are captured.
Pricing source: Kimi API Platform pricing index · checked 2026-05-28 · confidence low
Benchmark source: SWE-bench Leaderboards · checked 2026-06-02 · confidence medium
Benchmark-backed partial row only. Do not show price until exact Alibaba Cloud model pricing is captured.
Pricing source: Alibaba Cloud Model Studio pricing search result · checked 2026-05-28 · confidence low
Benchmark source: Aider LLM Leaderboards · checked 2026-05-28 · confidence medium
The provider or verified source did not publish this value.
This exact model was not found in the selected public benchmark source.
A source exists, but exact model alias, price mode, or context value was not verified.
The row has useful data, but not enough to support a strong recommendation.
This leaderboard does not combine SWE-bench, Aider, LiveCodeBench, arena scores, pricing tables, and usage signals into one fake universal score.
Pricing data prefers official provider pages. Aggregator or route-specific pricing must be labeled before it is used for a calculator prefill.
TTFT and tokens/sec only ship when a public source exists for the exact model or route. Otherwise speed is not disclosed.
A cheaper token price can become expensive if retries, output length, cache behavior, or failure cleanup change the workflow.
An AI coding model leaderboard compares language models for coding workflows such as coding agents, frontend generation, repo refactors, code review, bug fixing, and test generation. This page uses public benchmark evidence, pricing, context, caveats, and confidence labels instead of a single generic intelligence score.
There is no universal best LLM for coding agents. Start with models that have public coding benchmark evidence, source-backed API pricing, reliable context handling, and acceptable retry behavior for your workflow. Then estimate task cost before choosing a default model.
Those labels protect the user from fake precision. If an exact price, context window, speed signal, or benchmark result was not verified from a public source, the page says so instead of copying values from a similar model or making an assumption.
No. Real task cost depends on input length, output length, retries, tool calls, cache behavior, batch mode, failure rate, and human review time. Use benchmarks for evidence, then use task-cost assumptions for budgeting.