---
title: "The Gateway Charts, week 45 (Nov 2 to 8, 2026)"
series: "The Gateway Charts"
week: 2026-w45
period: 2026-11-02 to 2026-11-08 (UTC)
compared_with: 2026-10-26 to 2026-11-01
status: draft
license: CC BY 4.0
canonical: https://www.requesty.ai/data/weekly/2026-w45
json: https://www.requesty.ai/data/weekly/2026-w45/data.json
---

# The Gateway Charts, week 45: 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.

Period: Nov 2 to 8, 2026 (UTC), compared with Oct 26 to Nov 1, 2026. Publisher: Requesty.

## What to pay attention to

1. 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.
2. 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).
3. 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.
4. 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.
5. 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.
6. 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.
7. 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.

## Findings

### 1. 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.

- Value: 25.9% of tokens, each team capped at 2%
- Period: Nov 2 to 8, 2026 (UTC), compared with Oct 26 to Nov 1, 2026
- Why it matters: 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.
- What to do: See your own model mix by key and team (https://docs.requesty.ai/features/usage-analytics)
- Permalink: https://www.requesty.ai/data/weekly/2026-w45/top-model
- Social card: https://www.requesty.ai/data/weekly/2026-w45/top-model/square.png

### 2. 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).

- Value: 33.1% of tokens, each team capped at 2%
- Period: Nov 2 to 8, 2026 (UTC), compared with Oct 26 to Nov 1, 2026
- Why it matters: 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.
- What to do: Run Claude Code through Requesty (https://docs.requesty.ai/integrations/claude-code)
- Permalink: https://www.requesty.ai/data/weekly/2026-w45/coding-agents
- Social card: https://www.requesty.ai/data/weekly/2026-w45/coding-agents/square.png

### 3. 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.

- Value: 68.3% of active teams
- Period: Nov 2 to 8, 2026 (UTC), compared with Oct 26 to Nov 1, 2026
- Why it matters: 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.
- What to do: Switch models without changing your integration (https://docs.requesty.ai/quickstart)
- Permalink: https://www.requesty.ai/data/weekly/2026-w45/multi-model
- Social card: https://www.requesty.ai/data/weekly/2026-w45/multi-model/square.png

### 4. 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.

- Value: 87.5% of chat requests
- Period: Nov 2 to 8, 2026 (UTC), compared with Oct 26 to Nov 1, 2026
- Why it matters: 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.
- What to do: Get JSON you can parse with structured outputs (https://docs.requesty.ai/features/structured-outputs)
- Permalink: https://www.requesty.ai/data/weekly/2026-w45/defaults
- Social card: https://www.requesty.ai/data/weekly/2026-w45/defaults/square.png

### 5. 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.

- Value: 2.3× median time to first token, this week vs last
- Period: Nov 2 to 8, 2026 (UTC), compared with Oct 26 to Nov 1, 2026
- Why it matters: 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.
- What to do: Let the gateway pick the fastest host (https://docs.requesty.ai/features/latency-routing)
- Caveat: Latency is measured at the gateway, so it includes the host and the network, and reflects the prompts teams sent that week.
- Permalink: https://www.requesty.ai/data/weekly/2026-w45/speed-up
- Social card: https://www.requesty.ai/data/weekly/2026-w45/speed-up/square.png

### 6. 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.

- Value: 86.5% of requests on a fallback policy whose first host returned a provider error
- Period: Nov 2 to 8, 2026 (UTC), compared with Oct 26 to Nov 1, 2026
- Why it matters: 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.
- What to do: Set up a fallback policy (https://docs.requesty.ai/features/fallback-policies)
- Permalink: https://www.requesty.ai/data/weekly/2026-w45/rescues
- Social card: https://www.requesty.ai/data/weekly/2026-w45/rescues/square.png

### 7. 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.

- Value: 77.1% of input tokens
- Period: Nov 2 to 8, 2026 (UTC), compared with Oct 26 to Nov 1, 2026
- Why it matters: 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.
- What to do: Turn on auto caching (https://docs.requesty.ai/features/auto-caching)
- Permalink: https://www.requesty.ai/data/weekly/2026-w45/prompt-cache
- Social card: https://www.requesty.ai/data/weekly/2026-w45/prompt-cache/square.png

## Models by share of tokens

Unit: % of tokens, each team capped at 2%.

| Rank | Move | Name | Share | Change | Peak (12 wk) | Weeks on chart |
|---|---|---|---|---|---|---|
| 1 | = | deepseek-v4.1-flash (DeepSeek) | 25.9% | −3.93 pts | #1 | 9 |
| 2 | = | claude-opus-5-5 (Anthropic) | 10.1% | +2.27 pts | #1 | 7 |
| 3 | = | gpt-6-luna (OpenAI) | 8.39% | +0.84 pts | #3 | 7 |
| 4 | ▲ 1 | glm-5.3-flash (Z.ai) | 6.02% | +1.71 pts | #2 | 11 |
| 5 | ▼ 1 | gpt-6.1-sol (OpenAI) | 5.43% | −0.58 pts | #4 | 6 |
| 6 | = | claude-sonnet-5-5 (Anthropic) | 5.28% | +1.39 pts | #6 | 6 |
| 7 | ▲ 1 | deepseek-v4-flash (DeepSeek) | 4.29% | +0.84 pts | #1 | 12 |
| 8 | ▼ 1 | claude-haiku-5-5 (Anthropic) | 3.56% | −0.06 pts | #7 | 5 |
| 9 | ▲ 1 | nemotron-3-ultra-550b-a55b (NVIDIA) | 2.39% | −0.46 pts | #8 | 12 |
| 10 | ▲ 1 | gpt-5.6-luna (OpenAI) | 2.36% | −0.02 pts | #2 | 12 |
| 11 | ▼ 2 | claude-sonnet-5 (Anthropic) | 2.03% | −1.25 pts | #3 | 12 |
| 12 | ▲ 1 | nemotron-3.5-lightning-30b-a3b (NVIDIA) | 1.54% | +0.13 pts | #12 | 12 |
| 13 | ▲ 1 | gpt-6.1-terra (OpenAI) | 1.35% | −0.02 pts | #13 | 2 |
| 14 | ▲ 2 | gemini-3.8-flash (Google) | 1.30% | +0.36 pts | #7 | 9 |
| 15 | = | glm-5.3 (Z.ai) | 1.21% | −0.02 pts | #7 | 12 |
| 16 | ▲ 2 | ling-3.1-flash (Other) | 0.93% | +0.12 pts | #16 | 4 |
| 17 | ▲ 8 | claude-opus-5 (Anthropic) | 0.74% | +0.31 pts | #5 | 11 |
| 18 | ▼ 1 | deepseek-v4-pro (DeepSeek) | 0.73% | −0.20 pts | #4 | 11 |
| 19 | ▲ 9 | gemini-3.5-flash-lite (Google) | 0.57% | +0.27 pts | #17 | 10 |
| 20 | ▲ 3 | claude-sonnet-4-6 (Anthropic) | 0.53% | +0.06 pts | #10 | 12 |

## Apps by share of tokens

Unit: % of tokens, each team capped at 2%.

| Rank | Move | Name | Share | Change | Peak (12 wk) | Weeks on chart |
|---|---|---|---|---|---|---|
| 1 | = | OpenAI Python SDK (SDK or framework) | 20.6% | +1.09 pts | #1 | 11 |
| 2 | = | OpenCode (Coding agent) | 13.5% | −1.59 pts | #1 | 12 |
| 3 | = | Claude Code (Coding agent) | 11.5% | +0.71 pts | #2 | 12 |
| 4 | = | Vercel AI SDK (SDK or framework) | 9.36% | +0.83 pts | #3 | 12 |
| 5 | = | Pi (Coding agent) | 3.54% | −0.56 pts | #4 | 11 |
| 6 | ▲ 2 | DeepSeek Harness (Coding agent) | 2.72% | +0.66 pts | #6 | 11 |
| 7 | ▲ 3 | Anthropic SDK (SDK or framework) | 2.70% | +1.50 pts | #7 | 8 |
| 8 | ▼ 2 | OpenAI SDK (other) (SDK or framework) | 2.44% | −0.55 pts | #2 | 12 |
| 9 | = | LiteLLM (SDK or framework) | 1.93% | +0.50 pts | #9 | 10 |
| 10 | ▼ 3 | GitHub Copilot (Coding agent) | 1.81% | −0.90 pts | #5 | 12 |
| 11 | = | Kilo Code (Coding agent) | 1.27% | +0.42 pts | #7 | 12 |
| 12 | ▲ 1 | LangChain (SDK or framework) | 0.28% | +0.04 pts | #11 | 12 |

## Labs by share of tokens

Unit: % of tokens, each team capped at 2%.

| Rank | Move | Name | Share | Change | Peak (12 wk) | Weeks on chart |
|---|---|---|---|---|---|---|
| 1 | = | DeepSeek | 32.0% | −3.10 pts | #1 | 12 |
| 2 | = | Anthropic | 23.8% | +2.74 pts | #1 | 12 |
| 3 | = | OpenAI | 20.3% | +0.30 pts | #2 | 12 |
| 4 | = | Z.ai | 7.76% | +1.51 pts | #2 | 12 |
| 5 | ▲ 1 | Google | 4.14% | +0.39 pts | #4 | 12 |
| 6 | ▼ 1 | NVIDIA | 4.02% | −0.29 pts | #5 | 12 |
| 7 | = | MiniMax | 1.11% | −1.01 pts | #7 | 12 |
| 8 | ▲ 1 | Xiaomi | 0.96% | −0.09 pts | #6 | 11 |
| 9 | ▲ 1 | Moonshot | 0.87% | −0.17 pts | #6 | 12 |
| 10 | ▼ 2 | Mistral | 0.70% | −0.36 pts | #7 | 12 |

## Vital signs

| Metric | This week | Last week | Note |
|---|---|---|---|
| Input read from cache | 77.1% | 78.5% | Share of input tokens served from a prompt cache. |
| Input per output token | 46 : 1 | 40 : 1 | Tokens read for each token written. |
| Output spent thinking | 33.5% | 36.8% | Reasoning tokens as a share of output. |
| Turns ending in a tool call | 20.3% | 20.8% | Responses that asked for a tool instead of answering. |
| Median first token | 1.52 s | 1.57 s | Across every streaming request. |
| Median stream speed | 156 tok/s | 173 tok/s | Once the model starts talking. |
| Routed by a policy | 35.3% | 36.9% | Fallback, latency or load-balancing policies. |
| Routed by latency | 20.5% | 20.5% | 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.

## Cite this

Source: Requesty, The Gateway Charts, week 45 (Nov 2 to 8, 2026). https://www.requesty.ai/data/weekly/2026-w45

License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).
