Requesty
DraftFigures need sign-off before this week is published.

The Gateway Charts · Week 45, 2026

Last week in AI. Who climbed. Who slipped.

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.

Latest monthly release
October
2026
6 series
Explore →

01 Rankings

deepseek-v4.1-flash held #1 again.

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.

Share of tokens, each team capped at 2%Week 45 · Gateway data
This week, against last week
RankModelThis weekChange
  1. Rank 1. deepseek-v4.1-flash=DeepSeek25.9%−3.93 pts
  2. Rank 2. claude-opus-5-5=Anthropic10.1%+2.27 pts
  3. Rank 3. gpt-6-luna=OpenAI8.39%+0.84 pts
  4. Rank 4. glm-5.3-flash1Up 1 from #5Z.ai6.02%+1.71 pts
  5. Rank 5. gpt-6.1-sol1Down 1 from #4OpenAI5.43%−0.58 pts
  6. Rank 6. claude-sonnet-5-5=Anthropic5.28%+1.39 pts
  7. Rank 7. deepseek-v4-flash1Up 1 from #8DeepSeek4.29%+0.84 pts
  8. Rank 8. claude-haiku-5-51Down 1 from #7Anthropic3.56%−0.06 pts
  9. Rank 9. nemotron-3-ultra-550b-a55b1Up 1 from #10NVIDIA2.39%−0.46 pts
  10. Rank 10. gpt-5.6-luna1Up 1 from #11OpenAI2.36%−0.02 pts
Anthropic
claude-opus-5-5
This week
#2 · 10.1%
Last week
#2 · 7.81%
Change
+2.27 pts
Peak
#1
Weeks on chart
7 of 12
Data
Data for: deepseek-v4.1-flash held #1 again.
RankLast weekNameShareChange
11deepseek-v4.1-flash25.9%−3.93 pts
22claude-opus-5-510.1%+2.27 pts
33gpt-6-luna8.39%+0.84 pts
45glm-5.3-flash6.02%+1.71 pts
54gpt-6.1-sol5.43%−0.58 pts
66claude-sonnet-5-55.28%+1.39 pts
78deepseek-v4-flash4.29%+0.84 pts
87claude-haiku-5-53.56%−0.06 pts
910nemotron-3-ultra-550b-a55b2.39%−0.46 pts
1011gpt-5.6-luna2.36%−0.02 pts

02 Apps

Coding agents sent a third of tokens.

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).

Share of tokens by calling app, team-cappedWeek 45 · Gateway data
  • OpenAI Python SDKSDK or framework20.6%+1.09 pts
  • OpenCodeCoding agent13.5%−1.59 pts
  • Claude CodeCoding agent11.5%+0.71 pts
  • Vercel AI SDKSDK or framework9.36%+0.83 pts
  • PiCoding agent3.54%−0.56 pts
  • DeepSeek HarnessCoding agent2.72%+0.66 pts
  • Anthropic SDKSDK or framework2.70%+1.50 pts
  • OpenAI SDK (other)SDK or framework2.44%−0.55 pts
  • LiteLLMSDK or framework1.93%+0.50 pts
  • GitHub CopilotCoding agent1.81%−0.90 pts
Coding agent
OpenCode
This week
13.5%
Last week
15.1%
Change
−1.59 pts
Place in this chart
#2 of 10
Data
Data for: Coding agents sent a third of tokens.
NameThis weekLast week
OpenAI Python SDK20.6%19.5%
OpenCode13.5%15.1%
Claude Code11.5%10.8%
Vercel AI SDK9.4%8.5%
Pi3.5%4.1%
DeepSeek Harness2.7%2.1%
Anthropic SDK2.7%1.2%
OpenAI SDK (other)2.4%3.0%
LiteLLM1.9%1.4%
GitHub Copilot1.8%2.7%

03 Usage

Most teams used more than one model.

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.

Share of active teamsWeek 45 · Gateway data
How many each team used
Models68.3%use 2 or more
31.7%17.7%18.2%19.0%
Labs59.7%use 2 or more
40.3%23.6%16.7%
Providers61.7%use 2 or more
38.3%23.8%
Two or more
Share of active teams that mixed
Models
68.3%
Labs
59.7%
Providers
61.7%
Data
Data for: Most teams used more than one model.
1234 to 56 or moreTwo or more
Models31.7%17.7%13.3%18.2%19.0%68.3%
Labs40.3%23.6%16.7%14.1%5.2%59.7%
Providers38.3%23.8%14.8%12.7%10.5%61.7%

04 Usage

Most requests never set a temperature.

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.

Share of chat requests by settingWeek 45 · Gateway data
temperature
88of 100
leave it unset
reasoning_effort
78of 100
leave it unset
response_format
94of 100
ask for free text
One dot is one request in a hundred. Dots are rounded; the exact share is beside each value.
Data
Data for: Most requests never set a temperature.
ParameterValue sentShare of chat requests
temperatureNot set87.5%
temperatureSet to 08.4%
temperatureSet to 12.2%
temperatureSomething else1.9%
reasoning_effortNot set77.6%
reasoning_effortOn, no level16.7%
reasoning_effortLow to max4.8%
reasoning_effortOff or other0.9%
response_formatFree text94.4%
response_formatjson_object2.6%
response_formatjson_schema1.4%
response_formatOther1.6%

05 Speed

claude-sonnet-5 started 2.3× sooner.

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.

Median first token, streamingWeek 45 · Gateway data
Last weekThis week
ModelChange
  • claude-sonnet-52.3× sooner
  • claude-opus-51.4× sooner
  • gpt-6-luna1.2× sooner
  • claude-haiku-5-51.2× sooner
  • claude-sonnet-4-61.2× sooner
1 s3 s
Median first token, streaming
claude-sonnet-5
Last week
3.8 s
This week
1.69 s
Change
2.27× sooner
Relative
−56% Faster
Data
Data for: claude-sonnet-5 started 2.3× sooner.
ModelLast weekThis weekChange
claude-sonnet-53.8 s1.69 s2.3× sooner
claude-opus-54.3 s3.1 s1.4× sooner
gpt-6-luna1.87 s1.52 s1.2× sooner
claude-haiku-5-52.64 s2.19 s1.2× sooner
claude-sonnet-4-61.18 s992 ms1.2× sooner

06 Reliability

Most failed requests still got served.

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.

Requests whose first host failedWeek 45 · Gateway data
Routing policy
Served by another host 86.5%Still failed 13.5%
Fallback policy
86.5%
First host failed, still served
Latency policy
82.5%
First host failed, still served
Same model served
94.5%
Of rescues, same model on another host
Fallback policy
173 of 200 served
Served by another host
86.5%
Still failed
13.5%
One dot
0.5% of these requests
Data
Data for: Most failed requests still got served.
MeasureValue
Served by another host86.5%
Still failed13.5%
Fallback policy86.5%
Latency policy82.5%
Same model served94.5%

07 Caching

77.1% of input came from cache.

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.

Share of input tokens read from a prompt cacheWeek 45 · Gateway data
77.1%This week · 12-week range 72.5% to 82.2%
This week
Week of Nov 2
Input read from cache
77.1%
Against the week before
−1.4 pts
12-week average
76.8%
Data
Data for: 77.1% of input came from cache.
Week ofInput read from cache
2026-08-1780.4%
2026-08-2482.2%
2026-08-3177.3%
2026-09-0772.5%
2026-09-1475.5%
2026-09-2174.0%
2026-09-2876.2%
2026-10-0575.7%
2026-10-1274.8%
2026-10-1977.5%
2026-10-2678.5%
2026-11-0277.1%

Vital signs

How the traffic behaved

Input read from cache
77.1%
−1.40 pts
Input per output token
46 : 1
+6.10
Output spent thinking
33.5%
−3.30 pts
Turns ending in a tool call
20.3%
−0.50 pts
Median first token
1.52 s
−2.81%
Median stream speed
156 tok/s
−9.39%
Routed by a policy
35.3%
−1.60 pts
Routed by latency
20.5%
0.00 pts
Week 45

Full rankings

Models · 20Apps · 12Labs · 10Show
Models by share of tokensTop 20 · Week 45 · Gateway data
Models by share of tokens, each team capped at 2%, this week against last week
RankMoveModelThis weekChange
1=deepseek-v4.1-flashDeepSeek25.9%−3.93 pts
2=claude-opus-5-5Anthropic10.1%+2.27 pts
3=gpt-6-lunaOpenAI8.39%+0.84 pts
41Up 1 from #5glm-5.3-flashZ.ai6.02%+1.71 pts
51Down 1 from #4gpt-6.1-solOpenAI5.43%−0.58 pts
6=claude-sonnet-5-5Anthropic5.28%+1.39 pts
71Up 1 from #8deepseek-v4-flashDeepSeek4.29%+0.84 pts
81Down 1 from #7claude-haiku-5-5Anthropic3.56%−0.06 pts
91Up 1 from #10nemotron-3-ultra-550b-a55bNVIDIA2.39%−0.46 pts
101Up 1 from #11gpt-5.6-lunaOpenAI2.36%−0.02 pts
112Down 2 from #9claude-sonnet-5Anthropic2.03%−1.25 pts
121Up 1 from #13nemotron-3.5-lightning-30b-a3bNVIDIA1.54%+0.13 pts
131Up 1 from #14gpt-6.1-terraOpenAI1.35%−0.02 pts
142Up 2 from #16gemini-3.8-flashGoogle1.30%+0.36 pts
15=glm-5.3Z.ai1.21%−0.02 pts
162Up 2 from #18ling-3.1-flashOther0.93%+0.12 pts
178Up 8 from #25claude-opus-5Anthropic0.74%+0.31 pts
181Down 1 from #17deepseek-v4-proDeepSeek0.73%−0.20 pts
199Up 9 from #28gemini-3.5-flash-liteGoogle0.57%+0.27 pts
203Up 3 from #23claude-sonnet-4-6Anthropic0.53%+0.06 pts
Apps by share of tokensTop 12 · Week 45 · Gateway data
Apps by share of tokens, each team capped at 2%, this week against last week
RankMoveAppThis weekChange
1=OpenAI Python SDKSDK or framework20.6%+1.09 pts
2=OpenCodeCoding agent13.5%−1.59 pts
3=Claude CodeCoding agent11.5%+0.71 pts
4=Vercel AI SDKSDK or framework9.36%+0.83 pts
5=PiCoding agent3.54%−0.56 pts
62Up 2 from #8DeepSeek HarnessCoding agent2.72%+0.66 pts
73Up 3 from #10Anthropic SDKSDK or framework2.70%+1.50 pts
82Down 2 from #6OpenAI SDK (other)SDK or framework2.44%−0.55 pts
9=LiteLLMSDK or framework1.93%+0.50 pts
103Down 3 from #7GitHub CopilotCoding agent1.81%−0.90 pts
11=Kilo CodeCoding agent1.27%+0.42 pts
121Up 1 from #13LangChainSDK or framework0.28%+0.04 pts
Labs by share of tokensTop 10 · Week 45 · Gateway data
Labs by share of tokens, each team capped at 2%, this week against last week
RankMoveLabThis weekChange
1=DeepSeek32.0%−3.10 pts
2=Anthropic23.8%+2.74 pts
3=OpenAI20.3%+0.30 pts
4=Z.ai7.76%+1.51 pts
51Up 1 from #6Google4.14%+0.39 pts
61Down 1 from #5NVIDIA4.02%−0.29 pts
7=MiniMax1.11%−1.01 pts
81Up 1 from #9Xiaomi0.96%−0.09 pts
91Up 1 from #10Moonshot0.87%−0.17 pts
102Down 2 from #8Mistral0.70%−0.36 pts

Method

How we count

Week 45 covers Nov 2 to 8, 2026 (UTC), compared with Oct 26 to Nov 1, 2026.

  1. 01Aggregated from the Requesty production gateway for the week (Monday to Sunday, UTC), compared with the week before and the ten weeks before that.
  2. 02We publish shares, rates, ranks and latency only: no volumes, no customer data, no prompt content.
  3. 03In 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.
  4. 04An entry is named only when at least 20 teams used it and no single team is more than 60% of it.
  5. 05Latency 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.
  6. 06Internal test traffic, synthetic load and abusive accounts are excluded.
Source
Week 45 seed: the one-off Gateway Charts report (pre-pipeline) · scripts/weekly/seed/seed-2026-w45.py

For agents

Read this week as data

# Any week URL returns Markdown when asked for it
curl -H "Accept: text/markdown" \
  https://www.requesty.ai/data/weekly/2026-w45
Ask an AI about this weekChatGPT ↗Claude ↗Perplexity ↗
Source: Requesty, The Gateway Charts, week 45 (Nov 2 to 8, 2026). CC BY 4.0.

Next · Week 46

New charts every Monday