Claude Sonnet 5.5, GPT-6.1 Sol and Gemini 3.8 Flash are three proprietary models released within four weeks of each other in September 2026, and their prices, limits, tools and lifecycles differ in ways that matter when you pick one for production. This page sets the three side by side with identical fields for each: API ID, release date, context window, maximum output, knowledge cutoff, prices, thinking controls, tools, cloud availability, published safety findings, benchmarks, data handling and retirement status. It is written for developers, architects and buyers who need the documented facts in one place. Providers are listed alphabetically (Anthropic, Google, OpenAI) and order is never a ranking. The page gives facts only, with no verdict, and where a provider does not publish a fact the table says so.
Last checked: 2026-11-15.
This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider. Model details change, so confirm figures against each provider's current documentation, linked in the sources list.
Key takeaways
- Released within 27 days. Google released Gemini 3.8 Flash on 2 September 2026, Anthropic released Claude Sonnet 5.5 on 28 September 2026 and OpenAI released GPT-6.1 Sol on 29 September 2026.
- Two share a list price. Claude Sonnet 5.5 and GPT-6.1 Sol both list US$2 input and US$10 output per million tokens; Gemini 3.8 Flash lists US$0.75 and US$3.75 until 31 December 2026, doubling to US$1.50 and US$7.50 on 1 January 2027.
- About one million tokens in, different limits out. Anthropic states 1M tokens in and 128K out, OpenAI states 1,050,000 in and 128,000 out, and Google states 1,048,576 in and 65,536 out.
- Each provider controls reasoning differently. Sonnet 5.5 uses adaptive thinking with a default effort of high and cannot be set to disabled; GPT-6.1 Sol offers five effort levels with medium as default; Gemini 3.8 Flash offers low, medium and high thinking levels.
- No benchmark is published for all three under the same conditions. Anthropic and Google both report Terminal-Bench 4.0 (70.6% and 19.1%), but OpenAI's pages checked give no Terminal-Bench figure for GPT-6.1 Sol, and each provider compares against different rival models.
What are Claude Sonnet 5.5, Gemini 3.8 Flash and GPT-6.1 Sol?
All three are tiers below each provider's top model. Each is closed weight and proprietary, accepts text and images (Gemini 3.8 Flash also accepts audio, video and PDF) and returns text only.
- Claude Sonnet 5.5 (Anthropic): the Sonnet-line model in the Claude 5.5 family. Anthropic's models overview describes it as "The best combination of speed and intelligence" and rates its comparative latency as Fast. It sits between Claude Haiku 5.5 and Claude Opus 5.5. See the Claude Sonnet 5.5 profile.
- Gemini 3.8 Flash (Google): the Flash-tier model of the Gemini 3 family. Google's model page describes it as the most intelligent Flash model, designed for long-horizon software engineering, autonomous agents and complex enterprise workflows, with Flash speed and cost efficiency. The DeepMind model card states it builds on Gemini 3.7 Flash. See the Gemini 3.8 Flash profile.
- GPT-6.1 Sol (OpenAI): the Sol tier of OpenAI's GPT-6 generation, below GPT-6 Astra. OpenAI's model page describes it as "Near-Astra performance for complex work at a lower cost" and rates it Highest for reasoning and Fast for speed. See the GPT-6.1 Sol profile.
For each provider's full line-up, see Anthropic Models, Google Gemini Models and OpenAI Models.
Every model at a glance
As of 2026-11-15. Sources: Anthropic models overview, Google Gemini 3.8 Flash model page and model card, OpenAI gpt-6.1-sol model page.
| Field | Claude Sonnet 5.5 | Gemini 3.8 Flash | GPT-6.1 Sol |
|---|---|---|---|
| Provider | Anthropic | OpenAI | |
| API ID | claude-sonnet-5-5 | gemini-3.8-flash | gpt-6.1-sol |
| Release date | 28 September 2026 | 2 September 2026 | 29 September 2026 |
| Status | Active | Stable, generally available | Current (no deprecation listed) |
| Context window | 1M tokens | 1,048,576 tokens | 1,050,000 tokens |
| Max output | 128K tokens | 65,536 tokens | 128,000 tokens |
| Input types | Text, image | Text, image, video, audio, PDF | Text, image |
| Output types | Text | Text | Text |
| Knowledge cutoff | June 2026 (reliable cutoff) | March 2026 (model card) | 30 April 2026 |
| Reasoning control | Adaptive thinking, effort default high | Thinking levels low, medium, high | Effort low, medium (default), high, xhigh, max |
| Open weights | No | No | No |
| Parameters | Not stated by Anthropic | Not stated by Google | Not stated by OpenAI |
| Standard price, input / output per 1M tokens | US$2 / US$10 | US$0.75 / US$3.75 (to 31 Dec 2026) | US$2 / US$10 |
Google's model card notes that some knowledge domains may be limited to January 2025.
When was each model released?
As of 2026-11-15. Entries by date, oldest first. Sources: provider changelogs, release notes and announcements.
| Date | Event | Source |
|---|---|---|
| 2 September 2026 | Gemini 3.8 Flash generally available as gemini-3.8-flash on the Gemini API; same release date on Google's Gemini Enterprise Agent Platform; model card published | Gemini API changelog; Google Cloud model versions page; DeepMind model card |
| 28 September 2026 | Claude Sonnet 5.5 announced and released; system card published | Anthropic announcement; system card |
| 29 September 2026 | GPT-6.1 Sol released on the Responses and Chat Completions APIs; system card published | OpenAI API changelog; OpenAI deployment safety page |
| 8 October 2026 | Gemini API routes requests for gemini-3.7-flash to gemini-3.8-flash | Gemini API changelog |
| 8 October 2026 | Ultrafast service tier added for GPT-6.1 Sol in the Responses API | OpenAI API changelog |
For every 2026 launch across providers, see the LLM Release Tracker.
How much does each model cost?
As of 2026-11-15. US dollars per million tokens. Sources: Anthropic pricing page, Google Gemini API pricing page, OpenAI gpt-6.1-sol model page.
| Item | Claude Sonnet 5.5 | Gemini 3.8 Flash | GPT-6.1 Sol |
|---|---|---|---|
| Input | $2 | $0.75 ($1.50 from 1 Jan 2027) | $2.00 |
| Output | $10 | $3.75 ($7.50 from 1 Jan 2027), includes thinking tokens | $10.00 |
| Cached input (cache hit) | $0.10 | $0.075 ($0.15 from 1 Jan 2027) | $0.10 |
| Cache write or storage | $2.50 (5-minute write), $4 (1-hour write) | Storage $0.50 per 1M tokens per hour ($1.00 from 1 Jan 2027) | $2.50 (cache write) |
| Batch | $1 input / $5 output | $0.375 / $1.875 (Batch and Flex, half of Standard) | 50% lower than Standard (Batch and Flex) |
| Faster or priority tier | Not stated by Anthropic for this model on the pricing page checked | Priority: $1.35 / $6.75 | Fast mode 2x Standard; Ultrafast 6x Standard |
| Long-context rate | None: the 1M window is billed at standard rates | No separate long-context rate on the pricing page checked | Over 272K input tokens: 2x input and cache rates, 1.5x output, for the whole request |
| Regional or residency uplift | inference_geo "us": 1.1x on all token categories | Not stated by Google on the pricing page checked | 10% premium for regional processing where available |
| Web search grounding | $10 per 1,000 searches plus tokens | 5,000 free requests a month, then $14 per 1,000 | Not stated on the model page checked |
| Free API tier | Not stated by Anthropic on the pages checked | Free tier: input, output and caching free of charge | Free tier not supported |
Worked example: a request with 100,000 input tokens and 10,000 output tokens, at Standard rates with no caching or tools, costs US$0.30 on Claude Sonnet 5.5 ($0.20 + $0.10), US$0.30 on GPT-6.1 Sol ($0.20 + $0.10) and US$0.1125 on Gemini 3.8 Flash ($0.075 + $0.0375), rising to US$0.225 from 1 January 2027. A 400,000-token prompt changes the picture for GPT-6.1 Sol because it crosses the 272K threshold, while Anthropic states no long-context premium for Sonnet 5.5.
Effort settings drive output, and so cost, on all three. See Model Pricing Explained and How to set spend limits on LLM API keys.
What workloads does each provider position the model for?
As of 2026-11-15. Wording from each provider's own pages; this is positioning, not a test result.
| Model | Provider positioning | Source |
|---|---|---|
| Claude Sonnet 5.5 | "The best combination of speed and intelligence"; latency rated Fast | Anthropic models overview |
| Gemini 3.8 Flash | Long-horizon software engineering, autonomous agents and complex enterprise workflows, with Flash speed and cost efficiency | Google model page and changelog |
| GPT-6.1 Sol | "Near-Astra performance for complex work at a lower cost"; complex coding and professional work | OpenAI model page and API changelog |
Each provider also names its model as a migration target. Google's changelog tells new projects to use Gemini 3.5 Flash-Lite or 3.8 Flash. OpenAI's deprecations page names gpt-6.1-sol as the replacement for gpt-5, gpt-5-pro, o3 and o3-pro, all shutting down on 11 December 2026. Anthropic's announcement tells teams running Sonnet with thinking off to switch to the between_tools setting before moving to Sonnet 5.5.
What is the context window of each model?
As of 2026-11-15. Sources: provider model pages and pricing pages.
| Limit | Claude Sonnet 5.5 | Gemini 3.8 Flash | GPT-6.1 Sol |
|---|---|---|---|
| Input context | 1M tokens | 1,048,576 tokens | 1,050,000 tokens |
| Max output per response | 128K tokens | 65,536 tokens | 128,000 tokens |
| Price change with prompt length | None | None listed | Above 272K input tokens |
| Default context in provider products | Not stated by Anthropic on the pages checked | Not stated by Google on the pages checked | 500K default in Codex and ChatGPT Work |
See Context Windows Explained before relying on a full window.
Capabilities and tools compared
As of 2026-11-15. Features as listed on each provider's model page; "listed" means the provider names the feature for this model.
| Capability | Claude Sonnet 5.5 | Gemini 3.8 Flash | GPT-6.1 Sol |
|---|---|---|---|
| Reasoning settings | Adaptive thinking; effort default high; between_tools skips up-front thinking; disabled is rejected | Thinking levels low, medium, high; minimal returns an error | reasoning.effort low, medium (default), high, xhigh, max |
| Function or tool calling | Tool use documented; the ZDR page lists the MCP connector as a feature | Function calling listed | Function calling listed; tool calling requires the Responses API |
| Structured outputs | Listed on Anthropic's ZDR eligibility table | Listed | Listed |
| Web search | Web search tool priced on the pricing page | Search grounding and Google Maps grounding listed | Web search listed (Responses API) |
| Code execution | Code execution tool listed on the ZDR page (outside ZDR) | Code execution listed | Code interpreter and hosted shell listed |
| Computer use | Not stated on the Anthropic pages checked | Listed | Listed |
| Other listed tools | Citations, PDF support and prompt caching listed on the ZDR page | File search, URL context, caching, Batch API, Flex and Priority inference | File search, image generation, apply patch, skills, MCP, tool search |
| Not supported on this model | thinking type disabled | Audio generation, image generation, Live API | Audio and video input, fine-tuning |
Tools such as code execution, computer use and MCP let a model act with whatever credentials it is handed. Scope those credentials per tool and require approval for actions with side effects; see the Computer-Use and Browser Agents guide rather than a repeat here.
Where is each model available?
As of 2026-11-15. Sources: Anthropic announcement, Google model page and Google Cloud model versions page, OpenAI model page.
| Channel | Claude Sonnet 5.5 | Gemini 3.8 Flash | GPT-6.1 Sol |
|---|---|---|---|
| Provider's own API | Claude API, claude-sonnet-5-5 | Gemini API and Google AI Studio, gemini-3.8-flash | OpenAI API, gpt-6.1-sol |
| Amazon Web Services | Available (announcement) | Not stated by Google on the pages checked | Amazon Bedrock, openai.gpt-6.1-sol |
| Google Cloud | Available (announcement) | Gemini Enterprise Agent Platform, generally available | gpt-6.1-sol in private preview |
| Microsoft Azure | Available (announcement) | Not stated by Google on the pages checked | Not stated by OpenAI on the model page checked |
| Provider apps | Claude apps and Claude Code, default effort Medium; plan tiers not stated | Not stated by Google on the pages checked | Codex and ChatGPT Work, 500K default context; ChatGPT plan availability not stated |
Anthropic's announcement does not give cloud model IDs, and partner clouds set their own retirement schedules. The OpenAI changelog contains no entry stating that gpt-6.1-sol is available in ChatGPT consumer plans. For app-level plan details, see AI Chatbot Subscription Plans Compared.
Model cards and system cards
As of 2026-11-15. Only what each published document states.
| Field | Claude Sonnet 5.5 | Gemini 3.8 Flash | GPT-6.1 Sol |
|---|---|---|---|
| Document | System Card: Claude Sonnet 5.5 | Gemini 3.8 Flash model card (Google DeepMind) | GPT-6.1 Sol system card (OpenAI deployment safety page) |
| Published | 28 September 2026 | 2 September 2026 | 29 September 2026 |
| Safety framework | Responsible Scaling Policy | Frontier Safety Framework | Preparedness Framework |
| Framework finding | Treated as meeting CB-1 and Autonomy-1; does not cross any new RSP thresholds | Unlikely to reach any Tracked or Critical Capability Levels, by reference to the Gemini 3.7 Flash assessment | Critical in cybersecurity, High in biological and chemical, below High in AI self-improvement |
| Training data | Proprietary mix of public internet data, public and private datasets, permitted user data and synthetic data | Refers to the Gemini 3.7 Flash model card | Same types of data and training as GPT-6 Astra |
| Misuse safeguards stated | Three-stage cyber classifier system; blocked cyber requests fall back to Claude Sonnet 5 | Automated safety results reported against 3.7 Flash; child safety met launch thresholds in red teaming | Cyber access phased through Trusted Access for Cyber |
The three frameworks use different scales, so their levels are not equivalent. Anthropic's card states that cyber fallbacks happen automatically in Anthropic's apps but are opt-in on the API. See Model Cards and System Cards Explained and the Frontier AI Model Risk Frameworks Guide.
Benchmarks
As of 2026-11-15. Provider-reported only. A row appears only where at least two providers report the same benchmark version; independent leaderboard figures are not included because none were checked against their own pages for this comparison.
| Benchmark | Claude Sonnet 5.5 | Gemini 3.8 Flash | GPT-6.1 Sol | Source and date |
|---|---|---|---|---|
| Terminal-Bench 4.0 | 70.6% | 19.1% | Not stated by OpenAI on the pages checked | Anthropic announcement, 28 Sep 2026; DeepMind model card, Sep 2026 |
Other benchmarks share a name but not a version. Anthropic reports OSWorld 2.1 (partial) at 80.1% and Google reports OSWorld-2.0 (partial, batch tool enabled) at 59.0%; Anthropic reports GDPval-AA v2.1 at 1844 Elo and Google reports GDPVal-AA v2 at 1545 Elo; Anthropic reports Humanity's Last Exam with tools at 64.5% and Google reports HLE-Verified at 54.9%. These pairs are not comparable. OpenAI's GPT-6.1 Sol system card reports safety evaluations rather than these capability tests, for example SEC-Bench Pro pass@1 at 78.8% and ExploitGym at 35.1%.
| Model | Comparators in the provider's own table | Source |
|---|---|---|
| Claude Sonnet 5.5 | Claude Sonnet 5, Claude Opus 5.5, GPT-6 Sol | Anthropic announcement |
| Gemini 3.8 Flash | Gemini 3.7 Flash, Claude Opus 5, Claude Sonnet 5, GPT-5.6 Sol, GPT-5.6 Terra | DeepMind model card |
| GPT-6.1 Sol | GPT-6 Astra and GPT-6 Sol on safety evaluations | OpenAI system card |
None of the three providers compares its model with the other two models on this page. Scores depend on effort setting, tools and harness. See How to Read LLM Benchmarks.
Deprecations and retirements
As of 2026-11-15.
| Model or item | Details | Source |
|---|---|---|
| Claude Sonnet 5.5 | Active; not deprecated; tentative retirement not sooner than 28 September 2027; at least 60 days' notice for publicly released models | Anthropic model deprecations page |
| Gemini 3.8 Flash | Stable on the Gemini API with no deprecation listed; Google Cloud retirement 2 September 2027 or later, and listed dates "won't be moved to an earlier date" | Gemini API changelog; Google Cloud model versions page |
| GPT-6.1 Sol | Not listed as deprecated; named replacement for gpt-5, gpt-5-pro, o3 and o3-pro (shutdown 11 December 2026) | OpenAI deprecations page |
The Anthropic and Google dates are floors, not schedules. OpenAI publishes no date for GPT-6.1 Sol. Track all three in the LLM Deprecation and Retirement Tracker.
Data handling and security
As of 2026-11-15. API terms only; consumer app terms differ.
| Topic | Claude Sonnet 5.5 | Gemini 3.8 Flash | GPT-6.1 Sol |
|---|---|---|---|
| Training on API data | "Retained data is never used for model training without your express permission" | Paid-tier Standard content is not used to improve products; free-tier content may be | Not used to train or improve models by default since 1 March 2023 unless you opt in |
| Zero data retention | Available for Sonnet 5.5; enabled per organisation through Anthropic sales; covers eligible Messages and Token Counting features | Not stated on the pages checked | Zero data retention or Modified Abuse Monitoring by prior approval |
| Stated retention | Batch data 29 days; code execution up to 30 days; flagged content up to 2 years | Not stated on the pages checked | Abuse monitoring logs up to 30 days by default |
| Residency option | inference_geo "us" at 1.1x | Not stated on the pages checked | Regional processing at a 10% premium; Ultrafast supports US and EU residency |
The Google free-tier term is the one most likely to catch teams out: test keys on a free project can send data under different terms from production. Credentials matter as much as retention terms. See the Anthropic API Keys Security Guide, the Google Gemini API and AI Studio Security Guide, the OpenAI API Keys Security Guide and Workload Identity vs API Keys for LLM API Access.
How to choose a model: factual criteria
These criteria are documented facts to test against your workload, not a recommendation.
| If your requirement is | Facts to check |
|---|---|
| Lowest list price per token through 2026 | Gemini 3.8 Flash lists the lowest Standard rates, but they double on 1 January 2027; Sonnet 5.5 and GPT-6.1 Sol share $2 / $10 |
| Long prompts above 272K tokens | GPT-6.1 Sol applies 2x input and 1.5x output above 272K; Anthropic states standard rates across 1M; Google lists no long-context rate |
| Long outputs | 128K maximum output on Sonnet 5.5 and GPT-6.1 Sol; 65,536 on Gemini 3.8 Flash |
| Audio, video or PDF input | Gemini 3.8 Flash lists all three; Sonnet 5.5 and GPT-6.1 Sol list text and image input |
| Turning reasoning down | GPT-6.1 Sol lowest effort is low (none not supported); Gemini 3.8 Flash lowest is low (minimal errors); Sonnet 5.5 uses between_tools instead of disabled |
| Cloud you already use | Check the availability table: each model's listing on AWS, Google Cloud and Azure differs, and preview status applies to GPT-6.1 Sol on Google Cloud |
| Retention constraints | Sonnet 5.5 is available with ZDR; OpenAI ZDR needs approval; Google free-tier terms differ from paid |
| Planning horizon | Retirement floors of 28 September 2027 (Anthropic) and 2 September 2027 (Google Cloud); none published for GPT-6.1 Sol |
Larger organisations should also weigh which cloud contract the model sits under, since that sets data terms and billing. For coding-focused comparisons across more providers, see Coding LLMs Compared; for cheaper tiers, see Small and Low-Cost LLMs Compared.
Practitioner checklist
- Pin the exact ID (claude-sonnet-5-5, gemini-3.8-flash, gpt-6.1-sol) in code and record it in your model inventory with its retirement floor or status.
- Run your own evaluation on all three with matched effort or thinking settings; do not mix provider benchmark tables that use different rivals and versions.
- Model cost with caching, batch, long-context thresholds and residency multipliers, and add the 1 January 2027 Gemini 3.8 Flash price change to budgets.
- Remove unsupported settings before migrating: thinking disabled on Sonnet 5.5, minimal thinking on Gemini 3.8 Flash, none effort on GPT-6.1 Sol.
- Find any callers of gemini-3.7-flash on the Gemini API, which are already answered by Gemini 3.8 Flash, and any gpt-5 or o3 callers due to shut down on 11 December 2026.
- Confirm data terms per environment: ZDR enablement per organisation, OpenAI approval status, and that no production data passes through a Gemini free-tier project.
- Scope API keys or workload identities per application and set spend limits before enabling tools such as code execution or computer use.
Standards and references
- Anthropic: Models overview (accessed 2026-11-15)
- Anthropic: Pricing (accessed 2026-11-15)
- Anthropic: Claude Sonnet 5.5 announcement (accessed 2026-11-15)
- Anthropic: System Card, Claude Sonnet 5.5 (accessed 2026-11-15)
- Anthropic: Model deprecations (accessed 2026-11-15)
- Anthropic: Zero data retention (accessed 2026-11-15)
- Google: Gemini 3.8 Flash model page (accessed 2026-11-15)
- Google: Gemini API pricing (accessed 2026-11-15)
- Google: Gemini API changelog (accessed 2026-11-15)
- Google Cloud: Model versions and lifecycle (accessed 2026-11-15)
- Google DeepMind: Gemini 3.8 Flash model card (accessed 2026-11-15)
- OpenAI: GPT-6.1 Sol model page (accessed 2026-11-15)
- OpenAI: API changelog (accessed 2026-11-15)
- OpenAI: GPT-6.1 Sol system card (accessed 2026-11-15)
- OpenAI: Deprecations (accessed 2026-11-15)
- OpenAI: Data controls in the OpenAI platform (accessed 2026-11-15)
Related NHI Mgmt Group resources: Frontier AI Models Compared, Claude Sonnet Versions Compared, Gemini Flash and Flash-Lite Versions Compared, OpenAI Models.