GPT-6 Sol vs GPT-6.1 Sol
Seven-day successor — GPT-6.1 Sol scores higher on every benchmark and halves cache read costs, but runs 13% slower and drops Chat Completions tool calling.
OpenAI launched GPT-6 Sol on September 22, 2026 and superseded it with GPT-6.1 Sol at DevDay just seven days later. Both cost $2/$10 per MTok, but GPT-6.1 Sol brings a +4 Intelligence Index jump, +6.4 pp on DeepSWE, 50% cheaper cache reads — and trades speed for quality. This GPT-6 Sol vs GPT-6.1 Sol comparison covers every benchmark, the speed regression, API changes, and when the original Sol is still the right choice.
Last updated 2026-10-01 · Vendor data sourced from OpenAI Sol and OpenAI 6.1 Sol
At a Glance: GPT-6 Sol or GPT-6.1 Sol?
Stay on GPT-6 Sol when…
You depend on Chat Completions tool calling, use the none or minimal effort levels, or need the fastest possible output speed (74 t/s vs 64 t/s) for latency-critical applications.
Upgrade to GPT-6.1 Sol when…
You want higher benchmark scores across the board (+4 Intelligence Index, +6.4 pp DeepSWE), cheaper cache reads ($0.10 vs $0.20), and can tolerate a 13% speed drop and migrate to the Responses API for tool calling.
GPT-6 Sol vs GPT-6.1 Sol: What Changed
Every difference between the two models in one table — sourced from OpenAI's official documentation and Artificial Analysis.
| GPT-6 Sol | GPT-6.1 Sol | |
|---|---|---|
| Intelligence Index (max) | 48 | 52 (+4) |
| DeepSWE v1.1 (high) | 68.8% | 75.2% (+6.4 pp) |
| OSWorld 2.0 (max) | 64.4% | 71.4% (+7.0 pp) |
| ExploitBench (max) | 81.7% | 99.7% (+18.0 pp) |
| Cache read price | $0.20 / MTok | $0.10 / MTok (50% cheaper) |
| Output speed | 74 t/s | 64 t/s (13% slower) |
| TTFT (max effort) | 179.05s | 291.08s (63% slower) |
| Effort levels | none, minimal, low … max | low … max (dropped none/minimal) |
| Chat Completions tools | Supported | Not supported (Responses API only) |
| Knowledge cutoff | Apr 20, 2026 | Apr 30, 2026 |
The pattern is clear: GPT-6.1 Sol is smarter, cheaper to cache, but slower. Every benchmark improves; speed and some API features regress.
GPT-6 Sol vs GPT-6.1 Sol: Key Specifications
API identifiers, pricing, context limits and reasoning behavior — sourced from OpenAI's official documentation.
| Spec | GPT-6 Sol | GPT-6.1 Sol |
|---|---|---|
| API model ID | gpt-6-sol | gpt-6.1-sol |
| Released | Sep 22, 2026 | Sep 29, 2026 |
| Tagline | OpenAI's efficient reasoning model (superseded by GPT-6.1 Sol) | OpenAI's latest efficient reasoning model, successor to GPT-6 Sol |
| Latency | Fast | Fast |
| Input price (per MTok) | $2 | $2 |
| Output price (per MTok) | $10 | $10 |
| Context window | 1.05M tokens | 1.05M tokens |
| Max output | 128K tokens | 128K tokens |
| Thinking | Reasoning model, effort configurable | Reasoning model, effort configurable |
| Default effort | medium | medium |
| Knowledge cutoff | Apr 2026 | Apr 2026 |
| Input modalities | Text + images | Text + images |
| Retirement | Superseded by GPT-6.1 Sol (Sep 29, 2026) | At least 6 months notice per OpenAI policy |
GPT-6.1 Sol Pricing: Same Base, Cheaper Cache Reads
Base pricing is identical at $2/$10 per MTok. The only change: cache reads dropped from $0.20 to $0.10 per million tokens — a 50% reduction. Cache writes, batch pricing and fast mode are unchanged.
| Pricing Tier | GPT-6 Sol | GPT-6.1 Sol |
|---|---|---|
| Standard API (per MTok) | $2 / $10 | $2 / $10 |
| Batch API (50% off) | $1 / $5 | $1 / $5 |
| Cache write (5 min) | $2.50 | $2.50 |
| Cache write (1 hour) | $2.50 | $2.50 |
| Cache read | $0.20 (0.1x base) | $0.10 (0.05x base) |
For cache-heavy workloads, GPT-6.1 Sol saves on every cache hit. A workflow that caches 500K tokens per request saves $0.05 per call — meaningful at scale. For non-cached workloads, there is zero pricing difference.
Long-context surcharge remains the same on both: prompts exceeding 272K input tokens trigger 2x input/cache pricing and 1.5x output pricing.
Benchmark Results: GPT-6.1 Sol Leads Every Evaluation
Third-party and vendor benchmarks from Artificial Analysis and Kingy.
| Benchmark | GPT-6 Sol | GPT-6.1 Sol | Delta | Note |
|---|---|---|---|---|
| DeepSWE v1.1 | 68.8% | 75.2% | +6.4 pp | High effort; 6.1 matches Astra |
| OSWorld 2.0 (offline) | 64.4% | 71.4% | +7.0 pp | Max effort |
| AutomationBench 1.0.6 | 33.2% (xhigh) | 35.4% (medium) | +2.2 pp | 6.1 Sol at lower effort |
| ExploitBench | 81.7% | 99.7% | +18.0 pp | Max effort; near-perfect |
| Intelligence Index (max) | 48 | 52 | +4 | Artificial Analysis |
| Factual error rate (low) | 11.4% | 7.7% | -32% | Lower is better |
ExploitBench shows the largest jump: +18.0 pp to a near-perfect 99.7%. OSWorld gains +7.0 pp and DeepSWE gains +6.4 pp, bringing GPT-6.1 Sol to 75.2% — matching GPT-6 Astra's 74.8% at one-fifth of the per-task cost.
AutomationBench is notable: GPT-6.1 Sol scores 35.4% at medium effort vs Sol's 33.2% at xhigh — better results at lower effort means the effective cost per task drops even further.
Factual error rate drops 32% (11.4% to 7.7% at low effort), showing improved hallucination resistance even at the cheapest effort level.
Speed and Latency: GPT-6 Sol Is Faster Than GPT-6.1 Sol
Latency measurements from Artificial Analysis at max effort.
| Metric | GPT-6 Sol | GPT-6.1 Sol |
|---|---|---|
| Output speed | 74 tokens/s | 64 tokens/s |
| Time to first token | 179.05s | 291.08s |
GPT-6.1 Sol is 13% slower on output (64 t/s vs 74 t/s) and 63% slower on time to first token (291s vs 179s at max effort).
This is the main trade-off: GPT-6.1 Sol invests more compute per token to produce higher-quality results. For real-time chat or streaming UIs where first-token latency matters, the original GPT-6 Sol remains the better choice. For batch processing and background agents, the speed difference is usually irrelevant.
GPT-6.1 Sol API Changes: What Broke vs GPT-6 Sol
Chat Completions tool calling removed. GPT-6.1 Sol no longer supports function calling via the Chat Completions endpoint. Tool use is only available through the Responses API. If your codebase uses Chat Completions for tool calling, you must migrate before upgrading.
Effort levels reduced. GPT-6.1 Sol drops the “none” and “minimal” effort levels. The lowest setting is now “low.” GPT-6 Sol users who relied on none for function-calling-only requests (no reasoning overhead) need to switch to low effort, which adds some reasoning cost.
Multi-agent support (beta). GPT-6.1 Sol adds beta multi-agent support in the Responses API — a new capability not available on the original Sol.
GPT-6.1 Sol vs GPT-6 Astra: Near-Astra at Sol Pricing
OpenAI positions GPT-6.1 Sol as “near-Astra intelligence at one-fifth of Astra's price.” The numbers back this up:
| GPT-6.1 Sol | GPT-6 Astra | |
|---|---|---|
| DeepSWE v1.1 | 75.2% | 74.8% |
| Intelligence Index | ~52 | ~53 |
| DeepSWE cost/task | $0.65 | ~$3.25 |
| Input / output price | $2 / $10 | $10 / $50 |
GPT-6.1 Sol matches Astra on DeepSWE (75.2% vs 74.8%) and comes within 1 point on the Intelligence Index — at $2/$10 vs $10/$50 per MTok. For coding-heavy workloads, GPT-6.1 Sol is the clear value pick over Astra.
GPT-6 Sol vs GPT-6.1 Sol: Which to Choose by Use Case
Match your workload to the right model version.
| Scenario | Recommended | Why |
|---|---|---|
| Agentic coding | GPT-6.1 Sol | DeepSWE +6.4 pp (75.2% vs 68.8%); matches Astra at 1/5 the cost |
| Computer use / automation | GPT-6.1 Sol | OSWorld +7.0 pp; AutomationBench +2.2 pp at lower effort |
| Security research | GPT-6.1 Sol | ExploitBench +18 pp (99.7% vs 81.7%) |
| Cache-heavy workloads | GPT-6.1 Sol | Cache reads 50% cheaper ($0.10 vs $0.20/MTok) |
| Latency-critical streaming | GPT-6 Sol | 74 t/s vs 64 t/s; 179s vs 291s TTFT |
| Chat Completions tool calling | GPT-6 Sol | 6.1 Sol dropped Chat Completions tool support |
| Minimal-effort lightweight tasks | GPT-6 Sol | 6.1 Sol dropped none/minimal effort levels |
| Batch processing | GPT-6.1 Sol | Speed difference irrelevant; higher quality per dollar |
The Bottom Line
GPT-6.1 Sol is a strict benchmark upgrade — every evaluation score improves, cache reads are 50% cheaper, and it matches GPT-6 Astra on DeepSWE at one-fifth the cost. The trade-offs are real but narrow: 13% slower output, no Chat Completions tool calling, and two fewer effort levels. For most workloads, upgrading is the right call. Stay on the original GPT-6 Sol only if you need Chat Completions tools, the none/minimal effort levels, or the fastest possible output speed.
GPT-6 Sol vs GPT-6.1 Sol: FAQ
What are the main differences between GPT-6 Sol and GPT-6.1 Sol?
GPT-6.1 Sol is smarter but slower. It scores 52 vs 48 on the Intelligence Index (+4), gains 6.4 pp on DeepSWE and 7.0 pp on OSWorld, and halves cache read costs from $0.20 to $0.10 per MTok. But output speed drops from 74 to 64 tokens per second, time to first token nearly doubles, and Chat Completions tool calling is no longer supported. Base pricing stays the same at $2/$10 per MTok.
Is GPT-6.1 Sol a free upgrade from GPT-6 Sol?
Not entirely. Base pricing ($2/$10 per MTok) is identical and cache reads are 50% cheaper ($0.10 vs $0.20). But GPT-6.1 Sol is 13% slower on output speed and 63% slower on time to first token. It also drops the none and minimal effort levels and removes Chat Completions tool calling. If your workflow depends on any of these, the upgrade has trade-offs.
Why is GPT-6.1 Sol slower than GPT-6 Sol?
GPT-6.1 Sol trades speed for intelligence. Output drops from 74 to 64 tokens per second (13% slower) and time to first token rises from 179s to 291s at max effort. This is typical for model upgrades that increase reasoning depth — more compute per token produces better results at the cost of latency.
Should I upgrade from GPT-6 Sol to GPT-6.1 Sol?
Yes, for most workloads. GPT-6.1 Sol scores higher on every benchmark: DeepSWE +6.4 pp, OSWorld +7.0 pp, ExploitBench +18.0 pp. Cache reads are 50% cheaper. The only reasons to stay on GPT-6 Sol are: you need Chat Completions tool calling, you depend on the none or minimal effort levels, or you are latency-sensitive and cannot tolerate the 13% speed drop.
Does GPT-6.1 Sol support function calling?
Only via the Responses API. GPT-6.1 Sol dropped Chat Completions tool calling support. If your application uses the Chat Completions endpoint for function calling, you must migrate to the Responses API before upgrading, or stay on GPT-6 Sol.
How does GPT-6.1 Sol compare to GPT-6 Astra?
GPT-6.1 Sol matches Astra on DeepSWE (75.2% vs 74.8%) at roughly one-fifth of Astra's per-task cost. The Intelligence Index gap narrows to about 1 point (52 vs 53). GPT-6.1 Sol is positioned as near-Astra intelligence at Sol pricing — $2/$10 vs $10/$50 per MTok.
What effort levels does GPT-6.1 Sol support?
GPT-6.1 Sol supports low, medium (default), high, xhigh and max. It dropped the none and minimal levels that GPT-6 Sol had. If you used none for function-calling-only requests or minimal for extremely fast lightweight tasks, you need to adjust to at least low effort.
Is GPT-6 Sol being deprecated?
OpenAI has not announced a specific deprecation date. GPT-6 Sol's documentation now directs users to GPT-6.1 Sol, and OpenAI's deprecation policy guarantees at least 6 months notice. GPT-6 Sol remains accessible via the API but is effectively superseded.
How much cheaper are GPT-6.1 Sol cache reads?
Cache reads dropped from $0.20 to $0.10 per million tokens — a 50% reduction. This is the only pricing change between the two models. For cache-heavy workloads, GPT-6.1 Sol saves on every cache hit. All other pricing lines (input, output, cache write, batch, fast mode) remain identical.
When were GPT-6 Sol and GPT-6.1 Sol released?
GPT-6 Sol launched on September 22, 2026. GPT-6.1 Sol was announced at OpenAI DevDay on September 29, 2026 — just 7 days later. This is one of the fastest model supersessions in OpenAI's history.
Related Comparisons
- Claude Opus 5.5 vs GPT-6 Sol — cross-vendor comparison at the $2-$4/$10-$20 tier
- GPT-6 Luna vs GPT-6 Sol — OpenAI's budget tier (20x cheaper) vs Sol
- Claude Opus 5.5 vs GPT-6 Astra — flagship vs flagship ($4/$20 vs $10/$50)
Sources & References
- OpenAI GPT-6 Sol Model Card
- OpenAI GPT-6.1 Sol Model Card
- OpenAI API Pricing
- Artificial Analysis: GPT-6.1 Sol vs GPT-6 Sol
- BenchLM: GPT-6.1 Sol vs GPT-6 Sol
- Kingy: GPT-6.1 Sol Specs, Benchmarks & Task Costs
- Vellum: GPT-6.1 Sol Benchmarks Explained
- iWeaver: GPT-6.1 Sol vs GPT-6 Sol
- OpenRouter — live pricing: Sol / 6.1 Sol