{"series":"The Gateway Charts","canonical":"https://www.requesty.ai/data/weekly/2026-w45","license":"CC BY 4.0","citation":"Source: Requesty, The Gateway Charts, week 45 (Nov 2 to 8, 2026). https://www.requesty.ai/data/weekly/2026-w45","id":"2026-w45","year":2026,"week":45,"start":"2026-11-02","end":"2026-11-08","compare":{"start":"2026-10-26","end":"2026-11-01"},"weeks":["2026-08-17","2026-08-24","2026-08-31","2026-09-07","2026-09-14","2026-09-21","2026-09-28","2026-10-05","2026-10-12","2026-10-19","2026-10-26","2026-11-02"],"published":null,"status":"draft","title":"Last week in AI. Who climbed. Who slipped.","dek":"Every request through the Requesty gateway names a model, a host and the app that sent it. Each week we rank them against the week before and pull out what changed, and what it means for what you build.","findings":[{"id":"top-model","topic":"Rankings","claim":"deepseek-v4.1-flash held #1 on the Requesty gateway in week 45 (Nov 2 to 8, 2026) with 25.9% of tokens, down 3.93 points on the week before.","short":"deepseek-v4.1-flash held #1 again.","soWhat":"One model still carries a quarter of the traffic, but it gave up share for the first time since late September while claude-opus-5-5 gained 2.27 points. The top of the chart is less settled than the order suggests.","action":{"text":"See your own model mix by key and team","href":"https://docs.requesty.ai/features/usage-analytics"},"metric":{"value":25.93,"unit":"pct","denominator":"of tokens, each team capped at 2%","prev":29.86,"better":"up"},"footnote":"Share of tokens by model · Nov 2 to 8, 2026 vs Oct 26 to Nov 1","figure":{"kind":"ranking","ranking":"models","top":10,"highlight":"claude-opus-5-5"},"posts":{"x":["deepseek-v4.1-flash was the most used model on Requesty again last week (Nov 2 to 8): 25.9% of tokens. It gave up 3.93 pts, and claude-opus-5-5 picked up 2.27 pts to reach 10.1% at #2.","Top 3 models on Requesty last week, by share of tokens: deepseek-v4.1-flash 25.9%, claude-opus-5-5 10.1%, gpt-6-luna 8.4%. Same order as the week before.","One model still carries a quarter of the tokens on Requesty: deepseek-v4.1-flash, 25.9% last week (Nov 2 to 8). #1 in 7 of the last 8 weeks."],"linkedin":"deepseek-v4.1-flash held #1 on the Requesty gateway in week 45 (Nov 2 to 8, 2026) with 25.9% of tokens, down 3.93 points on the week before.\n\nOne model still carries a quarter of the traffic, but it gave up share for the first time since late September while claude-opus-5-5 gained 2.27 points. The top of the chart is less settled than the order suggests.\n\nThe Gateway Charts is our weekly read of what runs through the Requesty gateway: models, apps, speed and reliability, with the data behind every chart."},"url":"https://www.requesty.ai/data/weekly/2026-w45/top-model","cards":{"square":"https://www.requesty.ai/data/weekly/2026-w45/top-model/square.png","portrait":"https://www.requesty.ai/data/weekly/2026-w45/top-model/portrait.png","landscape":"https://www.requesty.ai/data/weekly/2026-w45/top-model/landscape.png"}},{"id":"coding-agents","topic":"Apps","claim":"Five coding agents (OpenCode, Claude Code, Pi, DeepSeek Harness and GitHub Copilot) sent 33.1% of tokens through the Requesty gateway in week 45 (Nov 2 to 8, 2026).","short":"Coding agents sent a third of tokens.","soWhat":"Agents now send almost as many tokens as the five biggest SDKs combined (37.0%). Their requests are long, cached and end in tool calls, so they set the shape of the bill and the latency budget.","action":{"text":"Run Claude Code through Requesty","href":"https://docs.requesty.ai/integrations/claude-code"},"metric":{"value":33.1,"unit":"pct","denominator":"of tokens, each team capped at 2%"},"footnote":"Share of tokens by calling app, team-capped · Nov 2 to 8, 2026","figure":{"kind":"bars","unit":"pct","rows":[{"label":"OpenAI Python SDK","value":20.58,"prev":19.49,"highlight":false,"sub":"SDK or framework"},{"label":"OpenCode","value":13.46,"prev":15.05,"highlight":true,"sub":"Coding agent"},{"label":"Claude Code","value":11.54,"prev":10.83,"highlight":true,"sub":"Coding agent"},{"label":"Vercel AI SDK","value":9.36,"prev":8.53,"highlight":false,"sub":"SDK or framework"},{"label":"Pi","value":3.54,"prev":4.1,"highlight":true,"sub":"Coding agent"},{"label":"DeepSeek Harness","value":2.72,"prev":2.06,"highlight":true,"sub":"Coding agent"},{"label":"Anthropic SDK","value":2.7,"prev":1.2,"highlight":false,"sub":"SDK or framework"},{"label":"OpenAI SDK (other)","value":2.44,"prev":2.99,"highlight":false,"sub":"SDK or framework"},{"label":"LiteLLM","value":1.93,"prev":1.43,"highlight":false,"sub":"SDK or framework"},{"label":"GitHub Copilot","value":1.81,"prev":2.71,"highlight":true,"sub":"Coding agent"}]},"posts":{"x":["About a third of all tokens on Requesty last week came from five coding agents: OpenCode 13.5%, Claude Code 11.5%, Pi 3.5%, DeepSeek Harness 2.7% and GitHub Copilot 1.8%. Together 33.1% (Nov 2 to 8).","OpenAI's Python SDK still sends the most tokens through Requesty (20.6% last week). Add up the five coding agents in the top 10 and you get 33.1%, close to the five SDKs and frameworks combined (37.0%).","Biggest climber among apps on Requesty last week: Anthropic SDK, #10 to #7. Five of the top 10 apps are now coding agents, sending 33.1% of tokens (Nov 2 to 8)."],"linkedin":"Five coding agents (OpenCode, Claude Code, Pi, DeepSeek Harness and GitHub Copilot) sent 33.1% of tokens through the Requesty gateway in week 45 (Nov 2 to 8, 2026).\n\nAgents now send almost as many tokens as the five biggest SDKs combined (37.0%). Their requests are long, cached and end in tool calls, so they set the shape of the bill and the latency budget.\n\nThe Gateway Charts is our weekly read of what runs through the Requesty gateway: models, apps, speed and reliability, with the data behind every chart."},"url":"https://www.requesty.ai/data/weekly/2026-w45/coding-agents","cards":{"square":"https://www.requesty.ai/data/weekly/2026-w45/coding-agents/square.png","portrait":"https://www.requesty.ai/data/weekly/2026-w45/coding-agents/portrait.png","landscape":"https://www.requesty.ai/data/weekly/2026-w45/coding-agents/landscape.png"}},{"id":"multi-model","topic":"Usage","claim":"68.3% of active teams on the Requesty gateway used two or more models in week 45 (Nov 2 to 8, 2026); 59.7% used two or more labs and 61.7% two or more providers.","short":"Most teams used more than one model.","soWhat":"Single-model stacks are the minority now. About one team in five ran six models or more, so switching and mixing models is routine work, not a migration project.","action":{"text":"Switch models without changing your integration","href":"https://docs.requesty.ai/quickstart"},"metric":{"value":68.3,"unit":"pct","denominator":"of active teams"},"footnote":"Share of active teams · Nov 2 to 8, 2026","figure":{"kind":"stacked","bins":["1","2","3","4 to 5","6 or more"],"rows":[{"label":"Models","values":[31.7,17.7,13.3,18.2,19],"headline":68.3,"headlineLabel":"use 2 or more"},{"label":"Labs","values":[40.3,23.6,16.7,14.1,5.2],"headline":59.7,"headlineLabel":"use 2 or more"},{"label":"Providers","values":[38.3,23.8,14.8,12.7,10.5],"headline":61.7,"headlineLabel":"use 2 or more"}]},"posts":{"x":["68.3% of active teams on Requesty used two or more models last week (Nov 2 to 8). 59.7% used two or more labs, and 61.7% two or more providers. About one in five ran six models or more.","Only 32% of teams on Requesty used a single model last week. The rest mix: 18% used two, 13% three, and 37% four or more. Plan for more than one model from day one.","6 in 10 teams on Requesty sent traffic to more than one lab last week, and 61.7% to more than one provider (fallbacks and routing policies count too). Switching models is normal now."],"linkedin":"68.3% of active teams on the Requesty gateway used two or more models in week 45 (Nov 2 to 8, 2026); 59.7% used two or more labs and 61.7% two or more providers.\n\nSingle-model stacks are the minority now. About one team in five ran six models or more, so switching and mixing models is routine work, not a migration project.\n\nThe Gateway Charts is our weekly read of what runs through the Requesty gateway: models, apps, speed and reliability, with the data behind every chart."},"url":"https://www.requesty.ai/data/weekly/2026-w45/multi-model","cards":{"square":"https://www.requesty.ai/data/weekly/2026-w45/multi-model/square.png","portrait":"https://www.requesty.ai/data/weekly/2026-w45/multi-model/portrait.png","landscape":"https://www.requesty.ai/data/weekly/2026-w45/multi-model/landscape.png"}},{"id":"defaults","topic":"Usage","claim":"87.5% of chat requests through the Requesty gateway in week 45 (Nov 2 to 8, 2026) never set a temperature; 77.6% left reasoning effort unset and 94.4% sent no response format.","short":"Most requests never set a temperature.","soWhat":"When a request leaves a setting out, the provider's default decides, and defaults differ by provider and change between model versions. Pin the settings your output depends on.","action":{"text":"Get JSON you can parse with structured outputs","href":"https://docs.requesty.ai/features/structured-outputs"},"metric":{"value":87.5,"unit":"pct","denominator":"of chat requests"},"footnote":"Share of chat requests by setting · Nov 2 to 8, 2026","figure":{"kind":"grid100","panels":[{"param":"temperature","headline":88,"phrase":"leave it unset","cats":[{"label":"Not set","pct":87.5,"dots":88},{"label":"Set to 0","pct":8.4,"dots":8},{"label":"Set to 1","pct":2.2,"dots":2},{"label":"Something else","pct":1.9,"dots":2}]},{"param":"reasoning_effort","headline":78,"phrase":"leave it unset","cats":[{"label":"Not set","pct":77.6,"dots":78},{"label":"On, no level","pct":16.7,"dots":16},{"label":"Low to max","pct":4.8,"dots":5},{"label":"Off or other","pct":0.9,"dots":1}]},{"param":"response_format","headline":94,"phrase":"ask for free text","cats":[{"label":"Free text","pct":94.4,"dots":94},{"label":"json_object","pct":2.6,"dots":3},{"label":"json_schema","pct":1.4,"dots":1},{"label":"Other","pct":1.6,"dots":2}]}]},"posts":{"x":["87.5% of chat requests on Requesty last week (Nov 2 to 8) never set a temperature. 77.6% left reasoning effort unset, and 94.4% sent no response format. Most apps run on the provider's defaults.","When a request on Requesty does set a temperature, it's usually 0: 8.4% of chat requests last week (Nov 2 to 8). 87.5% sent none at all, so the provider's default applied.","Structured output is rarer than you'd think. Last week 94.4% of chat requests on Requesty sent no response_format. If you parse model output in code, that's a lot of free text to trust."],"linkedin":"87.5% of chat requests through the Requesty gateway in week 45 (Nov 2 to 8, 2026) never set a temperature; 77.6% left reasoning effort unset and 94.4% sent no response format.\n\nWhen a request leaves a setting out, the provider's default decides, and defaults differ by provider and change between model versions. Pin the settings your output depends on.\n\nThe Gateway Charts is our weekly read of what runs through the Requesty gateway: models, apps, speed and reliability, with the data behind every chart."},"url":"https://www.requesty.ai/data/weekly/2026-w45/defaults","cards":{"square":"https://www.requesty.ai/data/weekly/2026-w45/defaults/square.png","portrait":"https://www.requesty.ai/data/weekly/2026-w45/defaults/portrait.png","landscape":"https://www.requesty.ai/data/weekly/2026-w45/defaults/landscape.png"}},{"id":"speed-up","topic":"Speed","claim":"claude-sonnet-5 reached its first token 2.3x sooner on the Requesty gateway in week 45 (Nov 2 to 8, 2026): a median of 1.69 s, down from 3.8 s the week before, on streaming requests.","short":"claude-sonnet-5 started 2.3× sooner.","soWhat":"First-token time moves week to week with the host, the model version and the prompts teams send. Routing on measured latency picks up a faster host as soon as it appears.","action":{"text":"Let the gateway pick the fastest host","href":"https://docs.requesty.ai/features/latency-routing"},"metric":{"value":2.3,"unit":"x","denominator":"median time to first token, this week vs last","better":"up"},"footnote":"Median first token, streaming · Nov 2 to 8 vs Oct 26 to Nov 1","figure":{"kind":"speedTape","rows":[{"name":"claude-sonnet-5","to":1691,"highlight":true,"from":3844},{"name":"claude-opus-5","to":3059,"highlight":false,"from":4336},{"name":"gpt-6-luna","to":1518,"highlight":false,"from":1873},{"name":"claude-haiku-5-5","to":2195,"highlight":false,"from":2639},{"name":"claude-sonnet-4-6","to":992,"highlight":false,"from":1180}]},"posts":{"x":["Speed-up of the week on Requesty: claude-sonnet-5. Median time to first token went from 3.8 s to 1.69 s (Nov 2 to 8 vs the week before), 2.3x sooner. Measured at the gateway on streaming requests.","claude-sonnet-5 sent its first token 2.3x sooner on Requesty last week: a median of 1.69 s, down from 3.8 s. A move like this blends the model, the host and the prompts, so we'll watch whether it holds.","The biggest first-token speed-up in our top 20 models last week (Nov 2 to 8): claude-sonnet-5, 3.8 s to 1.69 s at the median. Next: claude-opus-5, 4.3 s to 3.1 s."],"linkedin":"claude-sonnet-5 reached its first token 2.3x sooner on the Requesty gateway in week 45 (Nov 2 to 8, 2026): a median of 1.69 s, down from 3.8 s the week before, on streaming requests.\n\nFirst-token time moves week to week with the host, the model version and the prompts teams send. Routing on measured latency picks up a faster host as soon as it appears.\n\nThe Gateway Charts is our weekly read of what runs through the Requesty gateway: models, apps, speed and reliability, with the data behind every chart."},"caveats":["Latency is measured at the gateway, so it includes the host and the network, and reflects the prompts teams sent that week."],"url":"https://www.requesty.ai/data/weekly/2026-w45/speed-up","cards":{"square":"https://www.requesty.ai/data/weekly/2026-w45/speed-up/square.png","portrait":"https://www.requesty.ai/data/weekly/2026-w45/speed-up/portrait.png","landscape":"https://www.requesty.ai/data/weekly/2026-w45/speed-up/landscape.png"}},{"id":"rescues","topic":"Reliability","claim":"In week 45 (Nov 2 to 8, 2026), 86.5% of requests on a fallback policy whose first attempt hit a provider error were still served by another host on the Requesty gateway; 94.5% of those rescues served the same model.","short":"Most failed requests still got served.","soWhat":"A provider error only reaches your users when there is nowhere else to send the request. With a fallback policy, most failed first attempts were answered by the same model on another host.","action":{"text":"Set up a fallback policy","href":"https://docs.requesty.ai/features/fallback-policies"},"metric":{"value":86.5,"unit":"pct","denominator":"of requests on a fallback policy whose first host returned a provider error"},"footnote":"Requests whose first host failed · Nov 2 to 8, 2026","figure":{"kind":"split","total":200,"parts":[{"label":"Served by another host","pct":86.5,"state":"ok"},{"label":"Still failed","pct":13.5,"state":"failed"}],"stats":[{"label":"Fallback policy","value":86.5,"unit":"pct"},{"label":"Latency policy","value":82.5,"unit":"pct"},{"label":"Same model served","value":94.5,"unit":"pct"}]},"posts":{"x":["When a request's first host failed on Requesty last week, a fallback policy still got it served 86.5% of the time (Nov 2 to 8). Latency policies: 82.5%. 94.5% of rescues served the same model from another host.","A provider error doesn't have to reach your app. Last week 86.5% of requests whose first host failed were still served through a fallback policy on Requesty (Nov 2 to 8).","Rescues on Requesty last week: of requests that hit a provider error on the first try, 86.5% were served by the next host on a fallback policy. In 94.5% of rescues the answer came from the same model."],"linkedin":"In week 45 (Nov 2 to 8, 2026), 86.5% of requests on a fallback policy whose first attempt hit a provider error were still served by another host on the Requesty gateway; 94.5% of those rescues served the same model.\n\nA provider error only reaches your users when there is nowhere else to send the request. With a fallback policy, most failed first attempts were answered by the same model on another host.\n\nThe Gateway Charts is our weekly read of what runs through the Requesty gateway: models, apps, speed and reliability, with the data behind every chart."},"url":"https://www.requesty.ai/data/weekly/2026-w45/rescues","cards":{"square":"https://www.requesty.ai/data/weekly/2026-w45/rescues/square.png","portrait":"https://www.requesty.ai/data/weekly/2026-w45/rescues/portrait.png","landscape":"https://www.requesty.ai/data/weekly/2026-w45/rescues/landscape.png"}},{"id":"prompt-cache","topic":"Caching","claim":"77.1% of input tokens through the Requesty gateway in week 45 (Nov 2 to 8, 2026) were read from a prompt cache; models read 46 input tokens for every output token.","short":"77.1% of input came from cache.","soWhat":"Input is most of the bill for agent workloads, and the cache decides what that input costs. Compare models on what you pay after caching, not on the list price for output.","action":{"text":"Turn on auto caching","href":"https://docs.requesty.ai/features/auto-caching"},"metric":{"value":77.1,"unit":"pct","denominator":"of input tokens","prev":78.5,"better":"up"},"footnote":"Share of input tokens read from a prompt cache · Nov 2 to 8, 2026","figure":{"kind":"trend","unit":"pct","series":[{"label":"Input read from cache","values":[80.4,82.2,77.3,72.5,75.5,74,76.2,75.7,74.8,77.5,78.5,77.1],"highlight":true}]},"posts":{"x":["77.1% of input tokens on Requesty last week were read from a prompt cache (Nov 2 to 8). Models read 46 input tokens for every token they wrote. Output price is the smaller part of most bills.","For every token a model wrote on Requesty last week, it read 46. 77.1% of that input came from a prompt cache (Nov 2 to 8). Check your cache hit rate before you compare price lists.","Prompt caching on Requesty, 12 weeks: 80.4% of input tokens read from cache in mid-August, 77.1% last week (Nov 2 to 8). The workload is context, replayed on every turn."],"linkedin":"77.1% of input tokens through the Requesty gateway in week 45 (Nov 2 to 8, 2026) were read from a prompt cache; models read 46 input tokens for every output token.\n\nInput is most of the bill for agent workloads, and the cache decides what that input costs. Compare models on what you pay after caching, not on the list price for output.\n\nThe Gateway Charts is our weekly read of what runs through the Requesty gateway: models, apps, speed and reliability, with the data behind every chart."},"url":"https://www.requesty.ai/data/weekly/2026-w45/prompt-cache","cards":{"square":"https://www.requesty.ai/data/weekly/2026-w45/prompt-cache/square.png","portrait":"https://www.requesty.ai/data/weekly/2026-w45/prompt-cache/portrait.png","landscape":"https://www.requesty.ai/data/weekly/2026-w45/prompt-cache/landscape.png"}}],"rankings":[{"id":"models","label":"Models","unit":"pct","denominator":"of tokens, each team capped at 2%","rows":[{"rank":1,"last":1,"name":"deepseek-v4.1-flash","sub":"DeepSeek","share":25.93,"prevShare":29.86,"peak":1,"weeks":9,"series":[0,0,0,6.65,19.46,20.77,14.67,17.98,20.69,26.48,29.86,25.93],"ranks":[null,null,null,4,1,1,2,1,1,1,1,1]},{"rank":2,"last":2,"name":"claude-opus-5-5","sub":"Anthropic","share":10.08,"prevShare":7.81,"peak":1,"weeks":7,"series":[0,0,0,0,0,1.72,15.6,10.89,7.08,8.51,7.81,10.08],"ranks":[null,null,null,null,null,13,1,2,4,2,2,2]},{"rank":3,"last":3,"name":"gpt-6-luna","sub":"OpenAI","share":8.39,"prevShare":7.55,"peak":3,"weeks":7,"series":[0,0,0,0,0,5.05,10.09,9.97,8.29,7.52,7.55,8.39],"ranks":[null,null,null,null,null,6,4,3,3,4,3,3]},{"rank":4,"last":5,"name":"glm-5.3-flash","sub":"Z.ai","share":6.02,"prevShare":4.31,"peak":2,"weeks":11,"series":[0,3.52,9.92,11.87,12.21,10.42,10.72,7.67,6.96,5.53,4.31,6.02],"ranks":[null,8,2,2,3,3,3,5,5,5,5,4]},{"rank":5,"last":4,"name":"gpt-6.1-sol","sub":"OpenAI","share":5.43,"prevShare":6.01,"peak":4,"weeks":6,"series":[0,0,0,0,0,0,1.48,4.14,4.3,5.38,6.01,5.43],"ranks":[null,null,null,null,null,null,11,6,6,6,4,5]},{"rank":6,"last":6,"name":"claude-sonnet-5-5","sub":"Anthropic","share":5.28,"prevShare":3.89,"peak":6,"weeks":6,"series":[0,0,0,0,0,0,1.29,3.84,3.02,3.5,3.89,5.28],"ranks":[null,null,null,null,null,null,12,8,7,8,6,6]},{"rank":7,"last":8,"name":"deepseek-v4-flash","sub":"DeepSeek","share":4.29,"prevShare":3.45,"peak":1,"weeks":12,"series":[22.41,17.9,17.82,15.62,12.49,10.87,7.94,8.29,8.48,7.81,3.45,4.29],"ranks":[1,1,1,1,2,2,5,4,2,3,8,7]},{"rank":8,"last":7,"name":"claude-haiku-5-5","sub":"Anthropic","share":3.56,"prevShare":3.62,"peak":7,"weeks":5,"series":[0,0,0,0,0,0,0,0.84,1.69,2.45,3.62,3.56],"ranks":[null,null,null,null,null,null,null,18,11,10,7,8]},{"rank":9,"last":10,"name":"nemotron-3-ultra-550b-a55b","sub":"NVIDIA","share":2.39,"prevShare":2.85,"peak":8,"weeks":12,"series":[3.76,2.34,2.11,1.14,1.59,1.89,1.62,1.27,1.05,2.25,2.85,2.39],"ranks":[8,11,11,18,10,10,9,13,13,11,10,9]},{"rank":10,"last":11,"name":"gpt-5.6-luna","sub":"OpenAI","share":2.36,"prevShare":2.38,"peak":2,"weeks":12,"series":[10.39,8.68,9.61,9.04,9.8,5.2,3.77,2.88,2.79,2.07,2.38,2.36],"ranks":[2,3,3,3,4,5,7,9,8,12,11,10]},{"rank":11,"last":9,"name":"claude-sonnet-5","sub":"Anthropic","share":2.03,"prevShare":3.28,"peak":3,"weeks":12,"series":[7.33,5.48,7.03,5.05,3.88,6.09,4.73,4.01,2.76,2.45,3.28,2.03],"ranks":[3,4,4,5,6,4,6,7,9,9,9,11]},{"rank":12,"last":13,"name":"nemotron-3.5-lightning-30b-a3b","sub":"NVIDIA","share":1.54,"prevShare":1.41,"peak":12,"weeks":12,"series":[0.22,0.16,0.17,0.23,0.26,0.15,0.27,0.4,0.43,1.22,1.41,1.54],"ranks":[31,31,32,31,27,39,28,26,25,13,13,12]},{"rank":13,"last":14,"name":"gpt-6.1-terra","sub":"OpenAI","share":1.35,"prevShare":1.37,"peak":13,"weeks":2,"series":[0,0,0,0,0,0,0,0,0,0,1.37,1.35],"ranks":[null,null,null,null,null,null,null,null,null,null,14,13]},{"rank":14,"last":16,"name":"gemini-3.8-flash","sub":"Google","share":1.3,"prevShare":0.94,"peak":7,"weeks":9,"series":[0,0,0.67,1.16,1.39,1.07,0.85,2.53,6.77,4.44,0.94,1.3],"ranks":[null,null,19,17,12,17,17,10,null,7,16,14]},{"rank":15,"last":15,"name":"glm-5.3","sub":"Z.ai","share":1.21,"prevShare":1.23,"peak":7,"weeks":12,"series":[0.12,1.27,2.67,2.85,3.01,3.02,3.14,2.23,2.05,1.12,1.23,1.21],"ranks":[34,16,10,8,7,7,8,11,10,14,15,15]},{"rank":16,"last":18,"name":"ling-3.1-flash","sub":"Other","share":0.93,"prevShare":0.81,"peak":16,"weeks":4,"series":[0,0,0,0,0,0,0,0.1,0.22,0.32,0.81,0.93],"ranks":[null,null,null,null,null,null,null,42,34,26,18,16]},{"rank":17,"last":25,"name":"claude-opus-5","sub":"Anthropic","share":0.74,"prevShare":0.43,"peak":5,"weeks":11,"series":[7.3,5.3,3.76,3.9,6.2,2.43,1.15,0.81,0.5,0.55,0.43,0.74],"ranks":[null,5,7,6,5,8,14,19,20,20,25,17]},{"rank":18,"last":17,"name":"deepseek-v4-pro","sub":"DeepSeek","share":0.73,"prevShare":0.93,"peak":4,"weeks":11,"series":[5.76,6.5,2.71,1.55,0.75,1.27,1.08,1.05,0.8,0.85,0.93,0.73],"ranks":[4,null,9,14,17,16,15,14,15,15,17,18]},{"rank":19,"last":28,"name":"gemini-3.5-flash-lite","sub":"Google","share":0.57,"prevShare":0.3,"peak":17,"weeks":10,"series":[0.57,0.69,0.61,0.64,0.68,0.51,0.21,0.56,0.74,0.56,0.3,0.57],"ranks":[null,22,21,null,18,22,31,24,17,19,28,19]},{"rank":20,"last":23,"name":"claude-sonnet-4-6","sub":"Anthropic","share":0.53,"prevShare":0.47,"peak":10,"weeks":12,"series":[2.14,2.13,1.66,1.35,1.2,0.95,0.82,0.98,0.84,0.64,0.47,0.53],"ranks":[10,13,14,15,14,18,18,15,14,16,23,20]}]},{"id":"apps","label":"Apps","unit":"pct","denominator":"of tokens, each team capped at 2%","rows":[{"rank":1,"last":1,"name":"OpenAI Python SDK","sub":"SDK or framework","share":20.58,"prevShare":19.49,"peak":1,"weeks":11,"series":[21.54,23.25,24.14,20.68,22.06,21.85,21.38,20.79,18.8,19.17,19.49,20.58],"ranks":[1,null,1,1,1,1,1,1,1,1,1,1]},{"rank":2,"last":2,"name":"OpenCode","sub":"Coding agent","share":13.46,"prevShare":15.05,"peak":1,"weeks":12,"series":[12.06,10.88,13.13,13.17,11.93,15.93,11.29,10.26,11.44,12.3,15.05,13.46],"ranks":[2,1,2,2,2,2,2,2,3,2,2,2]},{"rank":3,"last":3,"name":"Claude Code","sub":"Coding agent","share":11.54,"prevShare":10.83,"peak":2,"weeks":12,"series":[8.64,6.44,6.19,8.8,4.53,6.61,7.09,9.61,11.95,11.5,10.83,11.54],"ranks":[4,4,5,4,6,4,4,3,2,3,3,3]},{"rank":4,"last":4,"name":"Vercel AI SDK","sub":"SDK or framework","share":9.36,"prevShare":8.53,"peak":3,"weeks":12,"series":[6.71,7.01,7.3,9.22,9.07,9.27,8.19,7.65,6.64,8.89,8.53,9.36],"ranks":[5,3,4,3,3,3,3,4,4,4,4,4]},{"rank":5,"last":5,"name":"Pi","sub":"Coding agent","share":3.54,"prevShare":4.1,"peak":4,"weeks":11,"series":[0,2.96,4.8,6.96,7.81,6.43,4.51,5.49,4.19,4.61,4.1,3.54],"ranks":[null,6,6,5,4,6,6,5,6,5,5,5]},{"rank":6,"last":8,"name":"DeepSeek Harness","sub":"Coding agent","share":2.72,"prevShare":2.06,"peak":6,"weeks":11,"series":[0.64,0.85,0.86,1.29,2,1.31,0.65,0.78,0.99,1.4,2.06,2.72],"ranks":[9,9,9,10,9,null,12,12,9,8,8,6]},{"rank":7,"last":10,"name":"Anthropic SDK","sub":"SDK or framework","share":2.7,"prevShare":1.2,"peak":7,"weeks":8,"series":[3.54,1.28,0.34,1.42,0.29,0.65,0.74,1.06,0.64,0.66,1.2,2.7],"ranks":[null,8,12,null,14,null,10,null,12,11,10,7]},{"rank":8,"last":6,"name":"OpenAI SDK (other)","sub":"SDK or framework","share":2.44,"prevShare":2.99,"peak":2,"weeks":12,"series":[11.74,9.52,7.84,4.77,5.54,6.57,3.24,2.87,2.87,2.71,2.99,2.44],"ranks":[3,2,3,6,5,5,7,7,7,6,6,8]},{"rank":9,"last":9,"name":"LiteLLM","sub":"SDK or framework","share":1.93,"prevShare":1.43,"peak":9,"weeks":10,"series":[1.27,0.82,0.45,0.55,0.98,0.74,0.78,1.57,0.98,1.35,1.43,1.93],"ranks":[null,null,10,12,11,12,9,9,10,9,9,9]},{"rank":10,"last":7,"name":"GitHub Copilot","sub":"Coding agent","share":1.81,"prevShare":2.71,"peak":5,"weeks":12,"series":[3.2,3.19,3.07,4.14,2.88,1.92,2,1.68,1.81,1.82,2.71,1.81],"ranks":[6,5,7,7,7,8,8,8,8,7,7,10]},{"rank":11,"last":11,"name":"Kilo Code","sub":"Coding agent","share":1.27,"prevShare":0.85,"peak":7,"weeks":12,"series":[1.54,1.81,1.4,1.98,1.12,1.01,0.62,1.14,0.91,1.03,0.85,1.27],"ranks":[7,7,8,8,10,9,13,11,11,10,11,11]},{"rank":12,"last":13,"name":"LangChain","sub":"SDK or framework","share":0.28,"prevShare":0.24,"peak":11,"weeks":12,"series":[0.34,0.41,0.44,0.38,0.3,0.25,0.26,0.27,0.2,0.21,0.24,0.28],"ranks":[11,11,11,13,13,15,14,14,16,14,13,12]}]},{"id":"labs","label":"Labs","unit":"pct","denominator":"of tokens, each team capped at 2%","rows":[{"rank":1,"last":1,"name":"DeepSeek","sub":"","share":31.98,"prevShare":35.08,"peak":1,"weeks":12,"series":[29.49,25.59,21.91,25.28,33.75,33.88,24.66,28.23,30.76,35.97,35.08,31.98],"ranks":[1,1,1,1,1,1,2,1,1,1,1,1]},{"rank":2,"last":2,"name":"Anthropic","sub":"","share":23.81,"prevShare":21.07,"peak":1,"weeks":12,"series":[21.82,18.75,18.37,17.53,15.03,14.6,26.51,24.35,18.36,20.2,21.07,23.81],"ranks":[2,3,3,3,4,3,1,2,3,2,2,2]},{"rank":3,"last":3,"name":"OpenAI","sub":"","share":20.28,"prevShare":19.98,"peak":2,"weeks":12,"series":[19.55,22.52,20.65,17.32,16.6,18.69,21.05,21.01,18.82,18.49,19.98,20.28],"ranks":[3,2,2,4,3,2,3,3,2,3,3,3]},{"rank":4,"last":4,"name":"Z.ai","sub":"","share":7.76,"prevShare":6.25,"peak":2,"weeks":12,"series":[5.73,9.91,16.14,17.77,17.08,14.57,14.49,10.35,9.53,6.92,6.25,7.76],"ranks":[6,4,4,2,2,4,4,4,5,5,4,4]},{"rank":5,"last":6,"name":"Google","sub":"","share":4.14,"prevShare":3.75,"peak":4,"weeks":12,"series":[9.3,9.36,7.13,7.23,7.03,6.06,3.22,5.06,9.67,7.5,3.75,4.14],"ranks":[4,5,5,5,5,5,5,5,4,4,6,5]},{"rank":6,"last":5,"name":"NVIDIA","sub":"","share":4.02,"prevShare":4.31,"peak":5,"weeks":12,"series":[6.39,5.17,4.39,3.4,2.77,2.58,2.71,3.56,3.05,3.75,4.31,4.02],"ranks":[5,6,7,8,6,8,6,6,6,6,5,6]},{"rank":7,"last":7,"name":"MiniMax","sub":"","share":1.11,"prevShare":2.12,"peak":7,"weeks":12,"series":[0.8,0.82,0.49,0.78,0.32,0.32,0.28,0.28,0.17,1.15,2.12,1.11],"ranks":[10,10,12,11,11,12,11,12,12,8,7,7]},{"rank":8,"last":9,"name":"Xiaomi","sub":"","share":0.96,"prevShare":1.05,"peak":6,"weeks":11,"series":[0.03,0.13,1.6,2.2,2.35,2.73,2.14,1.64,2.12,1.54,1.05,0.96],"ranks":[13,12,10,9,8,6,7,7,7,7,9,8]},{"rank":9,"last":10,"name":"Moonshot","sub":"","share":0.87,"prevShare":1.04,"peak":6,"weeks":12,"series":[3.91,3.69,4.57,3.72,1.44,2.66,1.3,1.05,0.95,0.46,1.04,0.87],"ranks":[7,7,6,6,9,7,8,9,9,10,10,9]},{"rank":10,"last":8,"name":"Mistral","sub":"","share":0.7,"prevShare":1.06,"peak":7,"weeks":12,"series":[0.9,1.88,2.36,3.41,2.43,1.2,0.6,0.69,0.68,0.79,1.06,0.7],"ranks":[9,8,8,7,7,9,10,10,10,9,8,10]}]}],"vitals":[{"id":"cached","label":"Input read from cache","metric":{"value":77.1,"unit":"pct","denominator":"of input tokens","prev":78.5,"better":"up"},"series":[80.4,82.2,77.3,72.5,75.5,74,76.2,75.7,74.8,77.5,78.5,77.1],"note":"Share of input tokens served from a prompt cache."},{"id":"io","label":"Input per output token","metric":{"value":46.1,"unit":"ratio","denominator":"input tokens per output token","prev":40},"series":[35.2,39.6,32.6,24.4,25.9,32.4,27.1,23.1,34.4,32.9,40,46.1],"note":"Tokens read for each token written."},{"id":"reason","label":"Output spent thinking","metric":{"value":33.5,"unit":"pct","denominator":"of output tokens","prev":36.8},"series":[37.3,40.1,42,33.9,16.3,21.8,21.1,36.3,28.1,24.2,36.8,33.5],"note":"Reasoning tokens as a share of output."},{"id":"tool","label":"Turns ending in a tool call","metric":{"value":20.3,"unit":"pct","denominator":"of responses","prev":20.8},"series":[27.9,31.5,22.5,17.5,14.4,20.9,13.6,17,18.6,17.8,20.8,20.3],"note":"Responses that asked for a tool instead of answering."},{"id":"ttft","label":"Median first token","metric":{"value":1524,"unit":"ms","denominator":"median, streaming requests","prev":1568,"better":"down"},"series":[877,685,824,743,929,1369,1517,1468,1943,1517,1568,1524],"note":"Across every streaming request."},{"id":"tps","label":"Median stream speed","metric":{"value":156.4,"unit":"tok_s","denominator":"median, streaming requests","prev":172.6,"better":"up"},"series":[85.1,73.8,85.4,90,103.9,112.9,132.7,142.4,164.2,160.4,172.6,156.4],"note":"Once the model starts talking."},{"id":"policy","label":"Routed by a policy","metric":{"value":35.3,"unit":"pct","denominator":"of tokens","prev":36.9,"better":"up"},"series":[21,14,30.7,37.7,36,40.9,37.3,33.6,29,35.2,36.9,35.3],"note":"Fallback, latency or load-balancing policies."},{"id":"latency_policy","label":"Routed by latency","metric":{"value":20.5,"unit":"pct","denominator":"of tokens","prev":20.5,"better":"up"},"series":[8.2,4.4,8.3,6.8,6.1,7.9,16.4,20.8,14.8,18,20.5,20.5],"note":"Tokens where the gateway picked the quickest host."}],"method":["Aggregated from the Requesty production gateway for the week (Monday to Sunday, UTC), compared with the week before and the ten weeks before that.","We publish shares, rates, ranks and latency only: no volumes, no customer data, no prompt content.","In the model, lab and app charts each team counts for at most 2% of a week's tokens, so one heavy team can't bend the chart.","An entry is named only when at least 20 teams used it and no single team is more than 60% of it.","Latency is measured at the gateway on streaming requests. It includes the host and the network, and reflects the prompts teams sent, not a controlled benchmark.","Internal test traffic, synthetic load and abusive accounts are excluded."],"sources":[{"label":"Week 45 seed: the one-off Gateway Charts report (pre-pipeline)","sql":"scripts/weekly/seed/seed-2026-w45.py"}]}