覆盖度 —— 计划 vs 落地
每一个「模型 × effort × 臂」都对照真实落地数据。批次是增量上线的 —— 这块板子如实画出「已有什么、还缺什么」。
数据截至 2026年9月5日 04:50414/ 555
75%已落地 · 占计划 555 席的 75%
414
已落地
4
部分
134
缺口
3
暂停
555
计划
已采集 1,800 个有效槽位
中国模型:官方 API 还有 26 个席位没跑,12 个家族目前零数据。
这是我们最想补上的缺口。这里每一个数字都由计划表对照真实落地数据算出,绝不是手写的 —— 点开下面任意一个缺口格子,就能自己出 key 把它跑出来。
17计划内家族26 / 39官方 api 空席12零数据家族70 / 118已有数据格
已落地— 每个变体都有 ≥1 个有效槽位部分— 有有效槽位,但不是每个变体都有跑过但空— config 已存在,但零有效槽位缺失— 还没有 config暂停— 计划内,但在 matrix 里被关掉
claude-fable-5claude
claude-haiku-4-5claude
claude-opus-4-8claude
claude-opus-5claude
claude-sonnet-5claude
deepseek-v4-flashdeepseek
highmax
deepseek-v4-prodeepseek
highmax
doubao-seed-2-1-pro-260628doubao
disabledenabled
ernie-5.1ernie
gemini-3.1-progemini
gemini-3.5-flashgemini
minimallowmediumhigh
gemini-3.6-flashgemini
minimallowmediumhigh
google/gemma-4-31b-itgemma
glm-5.2glm
gpt-5.2-codexgpt
lowmediumhighxhigh
gpt-5.4gpt
nonelowmediumhighxhigh
gpt-5.4-minigpt
nonelowmediumhigh
gpt-5.5gpt
high
gpt-5.6-lunagpt
gpt-5.6-solgpt
gpt-5.6-terragpt
grok-4.3grok
lowmediumhighdefault
grok-4.5grok
lowmediumhighdefault
hy3hunyuan
intern-s1-prointern
kwaipilot/kat-coder-pro-v2.5kat
kimi-k2.6kimi
offon
kimi-k2.7-codekimi
kimi-k3kimi
meta-llama/llama-4-maverickllama
LongCat-2.0longcat
mai-thinking-1mai
mercury-2mercury
mimo-v2.5-promimo
nonelowmediumhigh
MiniMax-M3minimax
mistral-medium-2604mistral
nvidia/nemotron-3-ultra-550b-a55bnemotron
amazon/nova-2-lite-v1nova
allenai/olmo-3.1-32b-thinkolmo
openpangu-2.0-flashpangu
default
qwen3.7-maxqwen
inclusionai/ring-2.6-1tring
spark-xspark
step-3.7-flashstep
gemini-3.1-pro-previewgemini
claude-fable-5claude
lowmediumhighxhighmax
claude-fable-5-1claude
lowmediumhighxhighmax
claude-haiku-4-5claude
lowmediumhighxhighmax
claude-opus-4-8claude
lowmediumhighxhighmax
claude-opus-5claude
lowmediumhighxhighmax
claude-sonnet-5claude
lowmediumhighxhighmax
gpt-6-astragpt
gpt-5.6-solgpt
gpt-5.6-terragpt
gpt-5.4gpt
lowmediumhighxhigh
gpt-5.4-minigpt
lowmediumhigh
gpt-5.5gpt
lowmediumhighxhigh
gpt-5.6-lunagpt
lowmediumhighxhigh
gpt-5.6-solgpt
lowmediumhighxhigh
gpt-5.6-terragpt
lowmediumhighxhigh
claude-fable-5multi
lowmediumhighxhighmax
claude-fable-5-thinkingmulti
lowmediumhighxhighmax
claude-opus-4-5multi
high
claude-opus-4-5-thinkingmulti
high
claude-opus-4-6multi
highmax
claude-opus-4-6-thinkingmulti
highmax
claude-opus-4-7multi
lowmediumhighxhighmax
claude-opus-4-7-thinkingmulti
lowmediumhighxhighmax
claude-opus-4-8multi
lowmediumhighxhighmax
claude-opus-4-8-thinkingmulti
lowmediumhighxhighmax
claude-opus-5multi
lowmediumhigh
claude-opus-5-thinkingmulti
lowmediumhighxhighmax
claude-sonnet-4multi
claude-sonnet-4-5multi
claude-sonnet-4-5-thinkingmulti
claude-sonnet-4-6multi
medium
claude-sonnet-4-6-thinkingmulti
medium
claude-sonnet-4-thinkingmulti
claude-sonnet-5multi
lowmediumhighxhighmax
claude-sonnet-5-thinkingmulti
lowmediumhighxhighmax
composer-2.5multi
cursor-grok-4.5multi
lowmediumhigh
cursor-grok-4.6multi
lowmediumhighxhigh
gemini-3-flashmulti
gemini-3.1-promulti
gemini-3.5-flashmulti
gemini-3.6-flashmulti
minimallowmediumhigh
glm-5.2multi
highmax
gpt-5-minimulti
gpt-5.1multi
lowhigh
gpt-5.2multi
lowhighxhigh
gpt-5.3-codexmulti
lowhighxhigh
gpt-5.4multi
lowmediumhighxhigh
gpt-5.4-minimulti
nonelowmediumhighxhigh
gpt-5.5multi
nonelowmediumhighextra-high
gpt-5.6-lunamulti
nonelowmediumhighxhighmax
gpt-5.6-solmulti
nonelowmediumhighxhighmax
gpt-5.6-terramulti
nonelowmediumhighxhighmax
kimi-k2.7-codemulti
kimi-k3multi
lowhighmax
claude-fable-5multi
lowmediumhighxhighmax
claude-fable-5-thinkingmulti
lowmediumhighxhighmax
claude-opus-4-5multi
claude-opus-4-5-thinkingmulti
high
claude-opus-4-6multi
high
claude-opus-4-6-thinkingmulti
highmax
claude-opus-4-7multi
mediumhighmax
claude-opus-4-7-thinkingmulti
lowhighxhigh
claude-opus-4-8multi
lowmediumhighxhighmax
claude-opus-4-8-thinkingmulti
lowmediumhighmax
claude-opus-5multi
lowmediumhigh
claude-opus-5-thinkingmulti
lowmediumhighxhigh
claude-sonnet-4multi
claude-sonnet-4-5multi
claude-sonnet-4-5-thinkingmulti
claude-sonnet-4-6multi
medium
claude-sonnet-4-6-thinkingmulti
medium
claude-sonnet-4-thinkingmulti
claude-sonnet-5multi
lowmediumhighxhighmax
claude-sonnet-5-thinkingmulti
lowmediumhighxhighmax
composer-2.5multi
cursor-grok-4.5multi
lowmediumhigh
cursor-grok-4.6multi
lowmediumhighxhigh
gemini-3-flashmulti
gemini-3.1-promulti
gemini-3.5-flashmulti
gemini-3.6-flashmulti
lowmedium
glm-5.2multi
high
gpt-5-minimulti
gpt-5.1multi
high
gpt-5.2multi
lowxhigh
gpt-5.3-codexmulti
lowhighxhigh
gpt-5.4multi
lowmediumxhigh
gpt-5.4-minimulti
nonemediumhigh
gpt-5.5multi
nonelowhighextra-high
gpt-5.6-lunamulti
nonemediumhighxhighmax
gpt-5.6-solmulti
nonelowmediumhighxhighmax
gpt-5.6-terramulti
nonelowmediumhighxhighmax
kimi-k2.7-codemulti
kimi-k3multi
lowhigh
deepseek-v4-flash-0731deepseek
offlowhighmax
deepseek-v4-flash-vision-expdeepseek
offlowhighmax
deepseek-v4-pro-0812deepseek
offlowhighmax
glm-5.2glm
deepseek-v4-flashmulti
deepseek-v4-promulti
glm-5multi
glm-5.1multi
glm-5.2multi
glm-5.3multi
grok-4.5multi
hy3-previewmulti
kimi-k2.5multi
kimi-k2.6multi
kimi-k2.7-codemulti
kimi-k3multi
mimo-v2-omnimulti
mimo-v2-promulti
mimo-v2.5multi
mimo-v2.5-promulti
minimax-m2.5multi
minimax-m2.7multi
minimax-m3multi
ox-alpha-freemulti
qwen3.5-plusmulti
qwen3.6-plusmulti
qwen3.7-maxmulti
qwen3.7-plusmulti
grok-4.5grok
lowmediumhigh
grok-4.6grok
lowmediumhighxhigh
kimi-k3kimi
claude-haiku-4-5multi
claude-opus-4-5multi
claude-opus-4-6multi
lowmediumhighmax
claude-opus-4-7multi
lowmediumhighxhighmax
claude-opus-4-8multi
lowmediumhighxhighmax
claude-sonnet-4multi
claude-sonnet-4-5multi
claude-sonnet-4-6multi
lowmediumhighmax
claude-sonnet-5multi
lowmediumhighxhighmax
deepseek-v3.2multi
glm-5multi
gpt-5.6-lunamulti
lowmediumhighxhighmax
gpt-5.6-solmulti
lowmediumhighxhighmax
gpt-5.6-terramulti
lowmediumhighxhighmax
minimax-m2.1multi
minimax-m2.5multi
qwen3-coder-nextmulti
mimo-v2.5-promimo
deepseek-v4-flashmulti
deepseek-v4-promulti
glm-5multi
glm-5.1multi
glm-5.2multi
glm-5.3multi
grok-4.5multi
hy3-previewmulti
kimi-k2.5multi
kimi-k2.6multi
kimi-k2.7-codemulti
kimi-k3multi
mimo-v2-omnimulti
mimo-v2-promulti
mimo-v2.5multi
mimo-v2.5-promulti
minimax-m2.5multi
minimax-m2.7multi
minimax-m3multi
ox-alpha-freemulti
qwen3.5-plusmulti
qwen3.6-plusmulti
qwen3.7-maxmulti
qwen3.7-plusmulti
qwen3.7-maxqwen
qwen3.8-max-previewqwen
qwen3.7-maxqwen
成本参考构建本数据集时实测的每次调用美元区间。一个 config = N 个有效槽位 × 2 个提示变体。
| 家族 | 臂 | 每次调用 USD | 依据 | 备注 |
|---|---|---|---|---|
| gpt | api | $0.05 – $2 | observed | 官方 /responses(background 模式)。5.4/5.4-mini/5.2-codex 全档均值约 $0.45/次;大模型 xhigh 档可达 ~$2/次。5.6 系尚未实测,预计更高。 |
| grok | api | $0.02 – $0.12 | observed | 官方 x.ai API:grok-4.3 + grok-4.5 全档 64 次调用合计约 $3.1。 |
| claude | CC | $0.01 – $0.6 | plan | Claude Code harness 走订阅;4 模型 × 5 档扫描的按量等效约 $64(fable-5 独占 ~$49)。订阅内零边际成本。 |
| claude | api | $0.05 – $1 | estimated | Anthropic 官方 API 尚未实跑;区间由 Claude Code 按量等效推算。 |
| gpt | codex-oauth | 订阅制 / 零成本 | plan | codex CLI 直登 ChatGPT 订阅;零边际成本。 |
| grok | grok-cli | 订阅制 / 零成本 | plan | grok CLI OAuth 订阅;零边际成本。 |
| multi | opencode | 订阅制 / 零成本 | plan | opencode go 订阅(22 模型,CLI harness + 网关 api);零边际成本。 |
| multi | go | 订阅制 / 零成本 | plan | opencode go 网关 api,同一订阅。 |
| deepseek | api | $0.001 – $0.03 | estimated | 官方第一方 API 牌价估算;尚未实跑。 |
| qwen | api | $0.002 – $0.05 | estimated | DashScope 第一方牌价估算;尚未实跑。 |
| glm | api | $0.002 – $0.05 | estimated | 智谱第一方牌价估算;尚未实跑。 |
| kimi | api | $0.002 – $0.05 | estimated | 月之暗面第一方牌价估算;尚未实跑。 |
| gemini | api | $0.01 – $0.3 | estimated | 尚未实跑(暂缓)。 |
实测区间 —— 随 effort 档位差异极大,仅作预算参考,并非报价。
点击任意「缺失」或「跑过但空」的格子,即可加入一个贡献批次。